Advanced Synthetic Aperture Radar (SAR) systems acquire a large volume of data, which necessitates the development of efficient data compression methods, beyond the current conventional techniques. Sentinel-1, as one the most popular SAR missions, provides global freely accessible data. However, the available raw data (i.e., Level-0 products) are quantized before being transferred, thus the statistics are different, hindering the validation of new algorithms mainly based on machine/deep learning paradigms. To enable elaboration of further SAR raw data compression, in this study, we propose a procedure to add random quantization noise to the decoded Sentinel-1 SAR raw data in order to obtain adapted uniformly quantized raw data that resemble the statistics of the uncompressed SAR raw data onboard the satellites. This method opens further opportunities to create large benchmarks for SAR raw data for data compression and other applications. The performance of data compression techniques (Block Adaptive Quantization (BAQ) and a complex-valued autoencoder-based data compression scheme) is evaluated on the adapted uniformly quantized raw data, and the effectiveness of the defined procedure is demonstrated.

In this work, the perspective of micro-Doppler analysis from synthetic aperture radar (SAR) images is assessed for large-scale areas. We propose a processing chain that implies a coarse vibrometry estimation based on azimuth sub-aperture decomposition and local Doppler centroid computations, followed by an in-depth analysis based on conventional SAR micro-Doppler algorithms. The methdology is assessed quantitatively using simulated data in keeping with Sentinel-1 IW imaging parameters.

The fusion of multimodal Remote Sensing (RS) data exploits the complementary nature of specific datasets, enhancing their analytical qualities beyond what is achievable when data are processed in isolation. Furthermore, integration of natively disparate data types, such as imagery and geospatial data, could substantially increase the image classification and semantic segmentation performances [1] [2].This study focuses on the automatically development of a multimodal benchmarking dataset, following the methodological principles of the BigEarthNet-MM [3] dataset and their integration in a minimal multimodal database. Unlike the original set, three new classes ("Flooded Area", "Burned Area", and "Volcanic Landscape") have been added in order to ensure a consistent performance evaluation using the same models as BigEarthNet-MM.The innovative aspect of the proposed dataset lies in its automatic developing and updating, highly accurate labelling-by-design process, and the existence of an additional data layer facilitating meaningful semantic connections as a prerequisite for a minimal multimodal database implementation.

In this work, the perspective synergy of Ground-Based Real Aperture Radar (GBRAR) measurements and micro-Doppler analysis of space-borne Synthetic Aperture Radar (SAR) images for the monitoring of vibration frequencies of large viaduct is discussed. A methodology for the processing GBRAR data and merging of ground-based and space-borne data is described. GBRAR data are interferometrically processed to provide time-range maps of Line-of-Sight (LoS) displacements. First results of GBRAR Ku-band measurements are presented. Vibration frequencies of different structural elements of the viaduct are derived. As a co-product, displacements of the bridge deck due to the crossing of vehicles are also obtained.

Sea ice plays a pivotal role in the Earth’s climate system and exhibits high sensitivity to shifts in temperature and atmospheric conditions. The precise and timely assessment of sea ice parameters is essential for comprehending and forecasting the climate changes. However, the vast volume of satellite data covering ice-covered regions is impractical to be subjectively assessed. Hence, the utilization of automated algorithms becomes mandatory to fully exploit the continuous data streams from satellites. In this paper, we propose a UNet transformer-based architecture, called UT-MHTA, to sea ice segmentation using SAR satellite imagery. Our UT-MHTA network replaces the conventional multi-head attention (MHA) block with a multi-head transposed attention (MHTA) which can capture long-range pixel interactions, while still remaining suitable for large images. Our method demonstrates superior performance compared to state-of-the-art methods, without drastically raising the computational complexity. In particular, UT-MHTA achieves a mean intersection over union (mIoU) of 68.76% on the AI4Arctic data set, with an inference time of 865ms for a 400 km2product.

The use of deep learning (DL) in Earth observation technology has become essential. Even though DL models do remarkably well in classifying high-resolution satellite images and extracting semantic information, they frequently need a lot of training data, which can be costly and time-consuming to obtain. With an emphasis on ocean synthetic aperture radar (SAR) image analysis, this research investigates the use of synthetically generated data using generative adversarial networks (GANs) for data augmentation. Relying on GAN-based generated images for remote sensing applications requires a thorough assessment of the quality and authenticity of the generated images, as well as validation of the model’s performance on real-world data. We assess the diversity and reliability of GAN-generated images by training a classification network on these images and evaluating their performance on real-world data. By comparing the classification accuracy in different experimental setups, we approximate the precision and recall for GANs performance.

Buildings represent pivotal entities in remote sensing imagery for various applications like urban planning and land resource management. Predominantly, methods for building footprint extraction in the literature focus on optical imagery with visual attributes that faithfully mirror the physical world. Nevertheless, the acquisition of high-quality optical images presents formidable challenges due to the susceptibility to illumination conditions and scene visibility. In contrast, synthetic aperture radar (SAR) images can be acquired in all-weather and all-time situations, unburdened by the aforementioned constraints. However, the coherent imaging mechanism engenders intricate complexities for building footprint extraction SAR images. To address this issue, this paper introduces the Boundary Information Distillation Network (BIDNet) to improve the prediction accuracy in SAR images by distilling knowledge from optical images. The proposed approach adopts a teacher-student framework, featuring two customized components: the Explicit Distillation Module (EDM) and the Latent Distillation Module (LDM). Different from the conventional practice of directly aligning feature maps, BIDNet focuses on leveraging the more conspicuous boundary information in optical images. The EDM operates by simultaneously yielding a boundary map to emphasize the boundary area and assimilating the explicit low-level features of two modalities. The LDM represents the structural attributes within the high-level latent feature space and aligns the representations of the two modalities. Within this module, intrinsic self-correlations among features originating from boundary regions are encoded, and so are the cross-correlations established between features from boundary regions and alternative areas. The two modules also serve as the conduit for knowledge distillation from the teacher network to the student network, enabling the utilization of optical imagery for enhancing the building footprint extraction in SAR imagery. Extensive experiments demonstrate that our BIDNet achieves state-of-the-art performance on the Multi-Sensor All Weather Mapping (MSAW) dataset, outperforming the strong baseline by 4.3-7.2 points in f1-score and 4.9-8.0 points in IoU. The source code and trained models will be publicly available.

The application of deep learning methods in various fields is rapidly increasing. The development of complex-valued (CV) networks that can process CV data has provided many opportunities for utilizing the immense capabilities of deep networks for CV data, including Synthetic Aperture Radar (SAR). However, the physical model and basic properties of the original SAR data must be preserved in the CV architecture. Without these properties, the physical parameters cannot be accurately retrieved from the SAR data. This study evaluates the competency of CV deep architectures to preserve the properties of the original SAR data and how it affects the retrieval of physical parameters. Ocean Surface Current (OSC) is an important parameter for ocean circulation and plays a vital role globally. In this work, the correlation Doppler estimation (CDE) method is used to estimate the OSC from SAR data before and after reconstruction with the CV autoencoder. The obtained OSCs are compared, and we demonstrate the ability of the CV deep architectures to learn the data model and preserve the original Doppler centroid (fDC) information in the SAR data. This research paves the way for the development of CV deep architectures for physical parameter retrieval and prediction from CV SAR data in future studies.

Recent advances in Synthetic Aperture Radar (SAR) sensors have enabled the acquisition of very high-resolution images with wide swaths, large bandwidth and in multiple polarization channels. As a result of the significant increase of SAR data size, an effective compression of the acquired data is of paramount importance. However, conventional data compression methods demonstrate limited effectiveness when applied to SAR data. In order to tackle this problem, in this study, a Complex-Valued (CV) end-to-end deep learning-based architecture based on convolutional autoencoders is proposed to compress Single Look Complex (SLC) SAR data. By relying on dual polarization SAR data, one of the polarization channels of the data is used as the side information to assist the reconstruction of the compressed channel with lower data loss. The obtained results demonstrate the remarkable potential and capability of CV deep learning-based methods for SAR data compression.

Advanced Differential SAR Interferometry and specifically Persistent Scatterers (PS) Interferometry have become popular techniques since they offer the possibility of large-scale analysis and monitoring of Earth's surface. In this work, we address the exploitation of widely used interferometric phase filtering method, SquueSAR, for the PS detection within a two-step processing scheme based on a sequence of low and high-resolution processing. We propose a modification of the iterative approach proposed in [1], in order to avoid phase shifts. The effects of the modification quantify in a significan improvement of the detection performances, verified on both simulated an real data.

Despite the permanent efforts to reduce emissions and achieve carbon neutrality a warmer climate is no longer to be avoided. The European mission „Adaptation to Climate Change" aims to build resilience by 2030 in at least 150 European communities and regions. At the same time, the „Destination Earth" (DestinE) initiative promotes the use of digital twins of the Earth enabling a thorough assessment of climate change by leveraging an accurate digital model of the Earth that can be used to monitor, model, and predict natural and human activity, and to develop and test scenarios for a more sustainable growth. Climate models describe changes at scales of 50km to 150km. However, adaptation measures shall be applied at human activities scales, from 10m to 1km. We propose to achieve this by scale-out novel paradigms of Artificial Intelligence for Earth Observation (AI4EO) including the use of coupled models across domains and spatiotemporal scales. The envisaged R&D work will be carried out in the project Competence Center for Climate Change Digital Twin Earth for forecasts and societal redressement: DTEClimate, in the frame of Romania National Recovery and Resilience Plan.

In this paper, an efficient routine for elliptical grid generation employed for the back-projection sped-up is proposed. The grid accommodates the particular bistatic setup formed by a space-born transmitter (Sentinel-l) and a ground-based stationary receiver (COBIS). Herein, the adapted elliptical grid is designed such that the computational complexity of the standard back-projection algorithm used for SAR image formation decreases. Specifically, the computational load is mitigated by reducing the density of the points in the cross-range direction. Such an elliptical grid leads to the formation of the SAR image in any arbitrary plane, preserving the range–azimuth (cross-range) spectrum of the final SAR image, making it suitable for further Doppler processing algorithms (e.g., common band selection in SAR interferometry).

Synthetic Aperture Radar (SAR) image generation using Generative Adversarial Networks (GANs) has gained significant attention in recent years. In addition, the ocean plays a crucial role in regulating Earth’s climate system. SAR images provide valuable information for ocean observation and analysis, aiding in the understanding of oceanic processes and their role in climate change. This study presents a GAN-based approach for generating realistic and diverse ocean pattern SAR images. The proposed methodology combines a style-based generator network with an adversarial discriminator network to learn and reproduce the complex and unique patterns present in SAR images. In order to avoid discriminator overfitting, which frequently occurs as a result of insufficient training data, an adaptive discriminator augmentation (ADA) mechanism has been exploited. By training the GAN with ADA, the generator learns to capture the spatial and statistical properties of oceanic phenomena in restricted data regimes. The experimental results demonstrate the effectiveness and potential of the proposed GAN-based approach for ocean pattern SAR image generation, opening new avenues for advancing ocean observation and analysis in the context of climate change mitigation.

Satellite image time series are a worthwhile source of information for a broad range of applications, especially in the context of future global challenges. The challenge to discover correlations, patterns or anomalies would be eased if the data analysts might benefit of visualization tools enabling them to grasp, briefly, the characteristics of the time series. Responding to these needs, this paper proposes a graphical user interface focused on SITS visualization. While overcoming the current software limitations we developed a Python GUI interface enabling remote sensing scientists who use Python for their research, to visually analyze the satellite image time series in the same environment.

A neuro-symbolic construct is proposed to provide a fast, trustworthy, and explainable way for detecting and monitoring the natural hazards, mainly based on remote sensing Earth Observation imagery but also supported by other kind of information. A combination of Graph Convolutional Network (GCN) and Convolutional Neural Network (CNN) models, together with some traditional programming methods and techniques - image processing algorithms and libraries, dedicated libraries for ontology building and programming interface, have been selected to support the implementation of a self-supervised ontology learning for natural hazard identification and monitoring. The specific use of CNN or GCN will depend on the specific tasks, but here we consider a hybrid approach that use both CNN and GCN, where CNN could be used to extract features from images which then serves as inputs to a GCN.

The Inverse Synthetic Aperture Radar (ISAR) provides images of objects that are rotating with respect to the radar. An efficient image focusing algorithm is required to generate ISAR imagery from the echoes of raw data. On the other hand, the dual-polarization decomposition technique enables precise retrieval of scattering mechanisms (H-α), allowing for various applications. In this paper, we propose a novel study case of 2D ISAR imaging of partial polarimetric data of natural targets. First, a stack of 2D complex-valued raw data with VV and VH polarizations is calibrated, and then the image focusing is applied using a match-filter and spherical-wave front compensation (SWFC) method. The eigenvector descriptors based decomposition is employed, and the scattering mechanism is identified using the Lee and Pottier H-α plane. To the best of the authors’ knowledge, ISAR images are used for the first time for this study. Given that decomposition enhances target characterization for studying scattering mechanisms, the application of the Radar Vegetation Index (RVI) demonstrates how dual-polarized ISAR images can be used for vegetation identification.

Knowledge systems in sustainable agriculture see a big gap with end users due to lack of easy-to-use interfaces with existing knowledge. Adding to the problem, decisions coming from black-box models are not understandable for most users. We try to bridge the gap with an integrated chain of explainable ML models to address the most useful applications in the agri-food industry. To make the integrated model available to users and help them draw benefits out of it, we also propose a novel idea of an explainable ML framework for interaction with human users. This human-in-the-loop approach makes ML models more trustworthy. End-users understand the output from ML models and also improve models with feedback. The application interface is also proposed to have features for multilingual communication among users to build communities. Feedback from communities help further refine ML models. The proposed system is named as Reusable Intelligent solution for Cultivation Enhancement (RISCE). In this article, we provide a demonstration of our system with an intrinsically explainable model for crop vigor analysis.

Compressive learning (CL) for Synthetic Aperture Radar (SAR) refers to the use of Compressive Sensing (CS) to reduce the amount of data required to represent SAR images while preserving key image features, with the goal of improving efficiency and lowering computational costs. In this paper, we propose a new, highly efficient RS technique based on creating a transcription between several classes. The proposed method is based on a novel CL theory, which is a joint signal processing and machine learning framework for inference from a signal that is represented by a small number of measurements obtained via linear projections of the data without first reconstructing the data. The results showed that, by minimizing the number of measurements or pixels in a data set, the accuracy curve will change depending on the data set and the method that is used. The algorithm reached an accuracy of about 80 % on SAR data, when using a SVM as classifier and a Binary sensing matrix when the number of pixels is reduced to 1/8 of the whole data.

Anomaly detection in Synthetic Aperture Radar (SAR) images is an important topic. However, the task is challenging due to the scarcity of anomalous samples and the lack of annotated data, which has led most algorithms in this field to be unsupervised. To address the issue, this article proposes a new loss that adds prior information. One of the main functions of an autoencoder is to reconstruct the input data as accurately as possible after encoding them in a latent vector. The proposed loss function guides the network using the Reed-Xiaoli (RX) detector and replaces any pixels in the input data deemed too abnormal with normal surrounding values. This approach incorporates a priori information in addition to the assumption that anomalies are largely under-represented compared to the rest of the image. An ablation study demonstrates that the proposed loss function improves detection performance.

Sea ice is a crucial component of the Earth’s climate system and is highly sensitive to changes in temperature and atmospheric conditions. Accurate and timely measurement of sea ice parameters is important for understanding and predicting the impacts of climate change. Nevertheless, the amount of satellite data acquired over ice areas is huge, making the subjective measurements ineffective. Therefore, automated algorithms must be used in order to fully exploit the continuous data feeds coming from satellites. In this paper, we present a novel approach for sea ice segmentation based on SAR satellite imagery using hybrid convolutional transformer (ConvTr) networks. We show that our approach outperforms classical convolutional networks, while being considerably more efficient than pure transformer models. ConvTr obtained a mean intersection over union (mIoU) of 63.68% on the AI4Arctic data set, assuming an inference time of 120ms for a 400×400 km2product.

In the context of accelerated urbanization, metropolitan green infrastructure is considered a strategic approach to ensure healthy and sustainable living environments. Earth Observation (EO) offers the right means for large scale and long term assessment and monitoring of such green areas and the entire urban environment. The methodology presented in this paper leverages one of the most common satellite missions for vegetation assessment, the Sentinel-2 mission, applies super-resolutions techniques to increase the image spatial resolution and quantifies the spectral radiation reflected by the ground in order to map the Earth’s biophysical properties. By considering multiple acquisitions over the same area, time series of spectral indices are generated and processed using LDA, a generative model well known for hierarchical latent information extraction in both text and image analysis. The resulting temporal signature of each topic is further correlated with the evidence of environmental indicators to underline the vegetation vulnerability and specificity of the species. A use case centered for the Bucharest city in Romania, was included.

In this paper, we briefly introduce the concept of physics guided learning for Synthetic Aperture Radar (SAR) and the potential advantages. Specifically, we propose a physics guided learning method for SAR airplane target feature representation, where the airplane scattering characteristics are extracted to guide the model training. To this end, the feature representation of physics guided learning is constrained to be aware of target scattering characteristics. The experimental results tentatively illustrate the effectiveness.

Recent studies revealed that the time-domain synthetic aperture radar (SAR) processors are more appropriate for future innovative SAR missions (e.g., ROSE-L, Harmony) not only due to their ability to form the SAR image on user-defined regions of interest (ROIs) but also for the straightforward accommodation to configurations wherein the azimuth spectrum folding occurs (e.g., TOPSAR). In this letter, we propose an accelerated Back-Projection (BP) SAR processor working in conjunction with a fast routine for generating the elliptical grid (laying on arbitrary planes) necessary for the sub-aperture based BP speed-up. The proposed workflow forms all the sub-aperture SAR images on the same coarse elliptical grid which before the coarse-to-fine grid interpolation are translated in the azimuth base-band. The fine-resolution SAR image is obtained by coherently integrating the sub-aperture images together with the concatenation of the corresponding fraction from the azimuth spectrum (unfolding) making the Single Look Complex (SLC) outcome proper for further Doppler-based processing. Our validation experiments indicate that the most suitable family of imaging planes is the one containing the receiver (fan-like grid) on bistatic scenarios with relatively large transmitter-receiver separation. The processing gain has been enhanced by one order of magnitude under low amplitude and phase distortions.

Sentinel-2 image super-resolution (SR) has proven advantageous in multiple data analysis pipelines, leading to a more comprehensive assessment of different environment-related metrics. This research aims to provide a method for super-resolving the 60-m bands provided by Sentinel-2 up to 10-m spatial resolution, using Gaussian process regression (GPR). While common GPR methods directly operate on raw data using carefully designed kernels, we propose a convolutional neural network (CNN)-based feature extraction kernel to directly process the input 10-m patches, applied in constructing the elements of the integrated covariance matrices. For each scene, a small number of training patches are sampled to optimize the CNN parameters and to construct the predictive mean function, the latter being further used for predicting super-resolved pixels for new input areas. We prove that our method is a reliable SR mechanism by assessing its performance both quantitatively, using metrics against other methods from literature, and qualitatively, through visual analysis of the results.

Deep learning models have achieved remarkable success in many different fields and attracted many interests. Several researchers attempted to apply deep learning models to Synthetic Aperture Radar (SAR) data processing, but it did not have the same breakthrough as the other fields, including optical remote sensing. SAR data are in complex domain by nature and processing them with Real-Valued (RV) networks neglects the phase component which conveys important and distinctive information. A Complex-Valued (CV) end-to-end deep network is developed in this study for the reconstruction and classification of CV-SAR data. Azimuth subaperture decomposition is utilized to incorporate physics-aware attributes of the CV-SAR into the deep model. Moreover, the correlation coefficient amplitude (Coherence) of the CV-SAR images depends on the SAR system characteristics and physical properties of the target. This coherency should be considered and preserved in the processing chain of the CV-SAR data. The coherency preservation of the CV deep networks for CV-SAR images, which is mostly neglected in the literature, is evaluated in this study. Furthermore, a large-scale CV-SAR annotated dataset for the evaluation of the CV deep networks is lacking. A semantically annotated CV-SAR dataset from Sentinel-1 Single Look Complex StripMap mode data (S1SLC_CVDL dataset) is developed and introduced in this study. The experimental analysis demonstrated the better performance of the developed CV deep network for CV-SAR data classification and reconstruction in comparison to the equivalent RV model and more complicated RV architectures, as well as its coherency preservation and physics-aware capability.

Spaceborne synthetic aperture radar (SAR) can provide accurate images of the ocean surface roughness day-or-night in nearly all weather conditions, being an unique asset for many geophysical applications. Considering the huge amount of data daily acquired by satellites, automated techniques for physical features extraction are needed. Even if supervised deep learning methods attain state-of-the-art results, they require a great amount of labelled data, which are difficult and excessively expensive to acquire for ocean SAR imagery. To this end, we use the subaperture decomposition (SD) algorithm to enhance the unsupervised learning retrieval on the ocean surface, empowering ocean researchers to search into large ocean databases. We empirically prove that SD improves the retrieval precision with over 20% for an unsupervised transformer auto-encoder network. Moreover, we show that SD brings an important performance boost when Doppler centroid images are used as input data, leading the way to new unsupervised physics guided retrieval algorithms.

Deep learning methods have become ubiquitous tools in many Earth observation applications, delivering state-of-the-art results while proving to generalize for a variety of scenarios. One such domain concerns the Sentinel-2 (S2) satellite mission, which provides multispectral images in the form of 13 spectral bands, captured at three different spatial resolutions: 10, 20, and 60 m. This research aims to provide a super-resolution mechanism based on fully convolutional neural networks (CNNs) for upsampling the low-resolution (LR) spectral bands of S2 up to 10-m spatial resolution. Our approach is centered on attaining good performance with respect to two main properties: consistency and synthesis. While the synthesis evaluation, also known as Wald’s protocol, has spoken for the performance of almost all previously introduced methods, the consistency property has been overlooked as a viable evaluation procedure. Recently introduced techniques make use of sensor’s modulation transfer function (MTF) to learn an approximate inverse mapping from LR to high-resolution images, which is on a direct path for achieving a good consistency value. To this end, we propose a multiobjective loss for training our architectures, including an MTF-based mechanism, a direct input–output mapping using synthetically degraded data, along with direct similarity measures between high-frequency details from already available 10-m bands, and super-resolved images. Experiments indicate that our method is able to achieve a good tradeoff between consistency and synthesis properties, along with competitive visual quality results.

This paper describes an image validation tool for the gen-eration of good quality Earth Observation (EO) benchmark datasets. We already developed an active learning based se-mantic annotation tool which allows user fast annotate images with few samples, the tool reached about 90% accuracy. A subsequent data cleaning tool then helps correct noisy data, thus increasing the number of correctly labeled images to be qualified as benchmark data. However, this work has an an-noying bottleneck, the manual correction via visual checks still costs a considerable amount of energy. Therefore, this paper aims to introduce an image validation tool and propose new metrics to distinguish different ambiguous cases within a dataset, based on pattern analysis of the data. The interactive visualization then enables users to visualize a dataset and explore unknown patterns. The benefits are two-fold: firstly, experiments show the proposed metrics greatly help decrease the manual labor whilst keeping the essential data, thus enhancing the de-gree of automation in the process of generating good quality benchmark datasets. Secondly, our approach provides possi-bilities to interactively visualize and explore very large-scale datasets in real time, thus providing help for further data mining.

Despite the state-of-the-art performance of the deep learning methods for Synthetic Aperture Radar (SAR) data classification, the Real-Valued (RV) networks neglect the phase component of the Complex-Valued (CV) SAR data and lose a lot of useful information. CV deep architectures have been developed in the recent years to exploit the amplitude and phase components of the CV data, in different fields. However, the superiority of CV models over RV models are proved to be different for each application, and more investigation into the advantages and disadvantages of implementing CV models for SAR data classification is necessary. In this study, the performance of the CV Convolutional Neural Network (CV-CNN) for Polarimetric SAR (PolSAR) data classification is compared with its RV equivalent network, in different contexts.

The Normalized Difference Vegetation Index (NDVI) is an important factor to be considered in vegetation tracking and analysis, which can be easily derived from multispectral (MS) images. However, the limitation imposed by the atmospheric conditions makes the calculation of this index difficult. Because of the clouds, only a limited number of multispectral bands can capture the land appropriately. Furthermore, the multispectral sensors are dependent on the sunlight, which makes the acquisition of data more limited. These limitations do not hinder other types of Earth Observation (EO) data, like the scenes captured by the Synthetic Aperture Radar (SAR). However, SAR images cannot be used in NDVI calculation. In this article, we propose a deep learning (DL) based method for NDVI estimation from SAR data. Using a database with corresponding MS and SAR patches, we calculate the NDVI for each sample, then use a convolutional neural network (CNN) for predicting the NDVI of SAR images. This simple method leads to a precision of 70% in NDVI estimation from SAR images.

The Sentinel-1 SAR images acquired using the TOPSAR modes i.e., IW and EW on cross-polarization are significantly affected by the thermal noise on low-back-scattering areas. For example, in the arctic and some desert zones both inter- swath and inter-burst noise amplification occurs. In this paper we propose a workflow for removing the thermal noise from Sentinel-1 ground detected SAR images on low back-scattering conditions by employing the co-polarization SAR image and the Gaussian Process Regression. Our processing flow uses the noise vectors provided in the European Space Agency (ESA) ground detected product and scales them such that a slightly over-denoised image is produced. Then, the Gaussian Process Regression is used to map the co-polarization SAR image into the cross-polarization SAR image. Prior to this step, a radiometric correction is applied on the co-polarization data, since its pixel values are heavily dependent on the incidence angle. Finally, the denoised cross-polarization image is obtained as a linear combination between the over-denoised version and the predicted image. Since, the co-polarization channel is employed for the prediction of the missing values in the cross-polarization channel there is no need for co-registration and the de noising procedure is trustworthy.

The resolution requirements of modern radar applications are increasing rapidly and cannot be fulfilled by the limited number of wide-band radar systems. Many approaches have been explored to solve this problem under the topic of super-resolution. In this paper, we propose a hybrid algorithm for resolution improvement, where we aim to combine the adaptability of deep neural networks with the reliability and expertise of traditional domain-specific SAR processing.

The spectral characteristics of single-look complex - inter-ferometric wide (SLC-IW) swath, terrain observation by progressive scan (TOPS), are significantly different from those of strip-map (SM). Due to the burst mode and series of sub-swaths, the target area is scanned for a short period of time. Therefore, swath width comes at the expense of azimuth resolution. To eliminate quadratic phase drift and achieve SLC baseband, significant processing is required. De-ramping is a necessary step to compute ocean circulation parameters. In this work, we extract ocean parameters from the complex echo signal based on data driven Doppler centroid$(f_{DC})$regardless of the OCN product information and geophysical$f_{DC}$image. The radial surface velocity (RSV) is retrieved from Doppler history, and the significant wave height (SWH) is estimated with an empirical relationship of RSV. The results of ocean circulation parameters are promising when compared with benchmark and in-situ data. This work demonstrates the efficacy and necessity of de-ramping the TOPS data for subsequent use in a variety of ocean remote sensing applications.

Super-resolution is used to recover high resolution images from low resolution images. We use this concept in a slightly different context to achieve higher quality knowledge from low resolution satellite images. The technique involves transfer learning from high to low resolution images using a Gaussian Process Regression model. We use high resolution drone images to train the model. This technique is applied in three case studies to verify the consistency of results in case of NDVI computation. However, the same technique can be applied to obtain for other application of satellite images in plant vigor assessment.

Causality is one of the most important topics in a Machine Learning (ML) research, and it gives insights beyond the dependency of data points. Causality is a very vital concept also for investigating the dynamic surface of our living planet. However, there are not many attempts for integrating a causal model in Remote Sensing (RS) methodologies. Hence, in this paper, we propose to use patch-based RS images and to represent each patch-based image by a single variable (e.g. entropy). Then we use a Structural Equation Model (SEM) to study their cause-effect relation. Moreover, the SEM is a simple causal model characterized by a Directed Acyclic Graph (DAG). Its nodes are causal variables, and its edges represent causal relationships among causal variables if and only if causal variables are dependent.

Machine Learning (ML) techniques are employed to analyze and process big Remote Sensing (RS) data, and one well-known ML technique is a Support Vector Machine (SVM). An SVM is a quadratic programming (QP) problem, and a D-Wave quantum annealer (D-Wave QA) promises to solve this QP problem more efficiently than a conventional computer. However, the D-Wave QA cannot solve directly the SVM due to its very few input qubits. Hence, we use a coreset ("core of a dataset") of given EO data for training an SVM on this small D-Wave QA. The coreset is a small, representative weighted subset of an original dataset, and any training models generate competitive classes by using the coreset in contrast to by using its original dataset. We measured the closeness between an original dataset and its coreset by employing a Kullback-Leibler (KL) divergence measure. Moreover, we trained the SVM on the coreset data by using both a D-Wave QA and a conventional method. We conclude that the coreset characterizes the original dataset with very small KL divergence measure. In addition, we present our KL divergence results for demonstrating the closeness between our original data and its coreset. As practical RS data, we use Hyperspectral Image (HSI) of Indian Pine, USA

Spaceborne synthetic aperture radar (SAR) can provide meters-scale images of the ocean surface roughness day-or-night in nearly all weather conditions. This makes it a unique asset for many geophysical applications. Sentinel-l SAR wave mode (WV) vignettes have made possible to capture many important oceanic and atmospheric phenomena since 2014. However, considering the amount of data provided, expanding applications requires a strategy to automatically process and extract geophysical parameters. In this study, we propose to apply subaperture decomposition (SD) as a preprocessing stage for SAR deep learning models. Our data-centring approach surpassed the baseline by 0.7%, obtaining state-of-the-art on the TenGeoP-SARwv data set. In addition, we empirically showed that SD could bring additional information over the original vignette, by rising the number of clusters for an unsupervised segmentation method. Overall, we encourage the development of data-centring approaches, showing that, data preprocessing could bring significant performance improvements over existing deep learning models.

Scene classification is one of the most important tasks in the remote sensing field. In general, remotely sensed data comprises targets of different nature with many detailed classes. Therefore, the classification of patches in a satellite scene is a challenging issue. To address the problem, the preferred alternative is to transform to polar coordinates and analyze angular distances. Prior works have so far considered angular distances between points, while ignoring that the target class is not a point, but a distribution. In this paper, we take advantage of this critical fact by using a point-to-probability distribution measure rather than an$\ell_{n}$norm. In this paper, two similarity measures (Euclidean and Mahalanobis) in two different feature space are experimentally investigated through some remote sensing datasets.

Domain-agnostic data retrieval has lately become essential amidst the availability of large-scale data from different types of sensors. However, the unavailability of a sufficient amount of samples of certain classes during training curtails the utility of existing retrieval models in remote sensing (RS) applications. Here, we propose a novel framework for zero-shot intermodal data retrieval of RS data. Thereupon, we design an encoder–decoder structure that ensures enhanced overlapping among the two data domains utilizing cross-triplet and cross-projection loss functions. Furthermore, we propose a sketch-based representation of the RS databaseEarth on Canvaswith diverse classes. We perform a thorough benchmarking of this data set and demonstrate that the proposed framework outperforms state-of-the-art methods for zero-shot sketch-based retrieval framework for RS data.

Graph convolution networks (GCNs) are useful in remote sensing (RS) image retrieval. It is found to be effective because, in a graph representation, the relative geometrical interactions between different regions (or segments) are appropriately captured, along with their region-wise features in their region adjacency graphs. Also, the attention mechanism has often been applied to the nodes to highlight the essential features in each node. In this regard, a significant amount of high-frequency information is missed since each image segment is effectively summarized within a single node. To account for this and increase the learning capacity, we propose to attend over the edge/adjacency matrix to highlight the interactions among meaningful regions that contribute to supervised learning from images. We exploit this novel edge attention mechanism together with node attention to highlight essential image context by allowing more importance to the meaningful neighboring regions that highlight a relevant node. We implement the proposed context-attended GCN framework for image retrieval on the benchmarked UC-Merced and the PatternNet datasets. We observe a notable improvement in the results compared to the state of the art.

Coastline extraction by exploiting optical images is challenging during adverse weather conditions. This letter proposes coastline extraction from synthetic aperture radar (SAR) data. Since collectingin-situdata is expensive and not always possible, the Doppler parameter is used to delineate coastlines when neitherin-situdata nor cloud-free optical images are available. We propose a novel coastline extraction method based on classic coastal dynamic variation, such as Doppler centroid (fDC), since coastline is static and has zero Doppler with respect to the dynamic sea-state. The results of the Doppler-based novel technique allow us to investigate the impact of natural hazards on coastline degradation. We compare the proposed method to state-of-the-art (SOA) coastline extraction methods based on polarimetric correlations and the reference method from Sentinel-2. The results show that using scattering from dual and cross-polarization for coastline extraction is more reliable than using co-polarization. Based on empirical distributions and using the constant false alarm rate (CFAR) method, the relevant threshold has been adapted to distinguish land and sea in an unsupervised manner. We compare the results of polarimetric and Sentinel-2 with Doppler-based coastline extraction, which emphasizes the accuracy of the proposedfDCmethod for extracting coastlines at full resolution.

Machine learning methods have shown tremendous success in understanding earth observation data; however, recently, there is a rising claim toward explainable machine learning approaches. Concerned researchers found interpretable visualizations to be greatly helpful in understanding how a model works. In this research, we propose a framework for interactive and interpretable visualization of remote sensing data using two machine learning models and an Elasticsearch (ES) database. Two explainable machine learning models, namely, bag-of-visual-words (BoVWs) and latent Dirichlet allocation (LDA) are chosen to model the data in an unsupervised manner and give a textual representation. The textualized remote sensing data are stored in an ES database. This framework offers several fast content-based search functionalities exploiting the full-text query capabilities of ES based on the respective representations and also offers an efficient storage mechanism for the data.

This letter studies how to program and assess a parameterized quantum circuit (PQC) for classifying Earth observation (EO) satellite images. In this exploratory study, we assess a PQC for classifying a two-label EO image dataset and compare it with a classic deep learning classifier. We use the PQC with an input space of only 17 quantum bits (qubits) due to the current limitations of quantum technology. As a real-world image for EO, we selected the Eurosat dataset obtained from multispectral Sentinel-2 images as a training dataset and a Sentinel-2 image of Berlin, Germany, as a test image. However, the high dimensionality of our images is incompatible with the PQC input domain of 17 qubits. Hence, we had to reduce the dimensionality of the input images for this two-label case to a vector with 16 elements; the 17th qubit remains reserved for storing label information. We employed a very deep convolutional network with an autoencoder as a technique for the dimensionality reduction of the input image, and we mapped the dimensionally reduced image onto 16 qubits by means of parameter thresholding. Then, we used a PQC to classify the two-label content of the dimensionally reduced Eurosat image dataset. A PQC classifies the Eurosat images with high accuracy as a classic deep learning method (and with even better accuracy in some instances). From our experiment, we derived and enhanced deeper insight into programming future gate-based quantum computers for many practical problems in EO.

In he past decade, object detection has achieved significant progress in natural images but not in aerial images, due to the massive variations in the scale and orientation of objects caused by the bird's-eye view of aerial images. More importantly, the lack of large-scale benchmarks has become a major obstacle to the development of object detection in aerial images (ODAI). In this paper, we present a large-scale Dataset of Object deTection in Aerial images (DOTA) and comprehensive baselines for ODAI. The proposed DOTA dataset contains 1,793,658 object instances of 18 categories of oriented-bounding-box annotations collected from 11,268 aerial images. Based on this large-scale and well-annotated dataset, we build baselines covering 10 state-of-the-art algorithms with over 70 configurations, where the speed and accuracy performances of each model have been evaluated. Furthermore, we provide a code library for ODAI and build a website for evaluating different algorithms. Previous challenges run on DOTA have attracted more than 1300 teams worldwide. We believe that the expanded large-scale DOTA dataset, the extensive baselines, the code library and the challenges can facilitate the designs of robust algorithms and reproducible research on the problem of object detection in aerial images.

The performance of a deep-learning-based model primarily relies on the diversity and size of the training dataset. However, obtaining such a large amount of labeled data for practical remote sensing applications is expensive and labor-intensive. Training protocols have been previously proposed for few-shot learning (FSL) and zero-shot learning (ZSL). However, FSL is not compatible with handling unobserved class data at the inference phase, while ZSL requires many training samples of the seen classes. In this work, we propose a novel training protocol for image retrieval and name it aslabel-deficit zero-shot learning(LDZSL). We use this novel LDZSL training protocol for the challenging task of cross-sensor data retrieval in remote sensing. This protocol uses very few labeled data samples of the seen classes during training and interprets unobserved class data samples at the inference phase. This strategy is critical as some data modalities are hard to annotate without domain experts. This work proposes a novel bi-level Siamese network to perform the LDZSL cross-sensor retrieval of multispectral and SAR images. We utilize the available geo-referenced SAR and multispectral data to domain align the embedding features of the two modalities. We experimentally demonstrate the proposed model’s efficacy using the So2Sat dataset compared to the existing state-of-the-art models of the ZSL framework trained under a reduced training set. We also show the generalizability of the proposed model using a sketch-based image retrieval task. Experimental results on the Earth on Canvas dataset exhibit comparative performance over the literature.

In the past decade, anomaly detection has experienced an expanding attraction in satellite data analysis. Monitoring wildfire dynamics plays a substantial part in global land management, i.e., to detect and determine the expansion of such areas, estimate the deterioration of forest regions, and assist the intervention plan. In this paper, we proposed an approach for Sentinel-2 scenes that detects anomalies in burned area contexts using a rank-ordered method on a single post-event image. We adopted a self-supervised paradigm in learning image representations by training a deep convolutional model to differentiate between a series of geometric transformations. Dirichlet distributions are selected as priors to characterize the variability of random multinomial distribution in multispectral data. Dirichlet precision parameters are computed from observed data and are used to construct a ranking function that quantifies the degree of anomaly in data based on softmax responses given by the trained classifier. We evaluated the performance of the proposed method on a cumulative effort of two remote sensing tasks, open-set detection (i.e., test datasets contain classes unseen at the training time) and location separation (i.e., test datasets include images from distinct spatial location than the training images). The experiments were performed on three different datasets, BigEarthNet and two actual burned area Sentinel-2 datasets, i.e., from predisposed zones to fire events, Australia, respectively Bolivia. We retained in all test datasets state-of-the-art performance, considering the substantial and diversified types of natural anomalies in multispectral data.

This article introduces a compressive sensing (CS)-based approach for increasing bistatic synthetic aperture radar (SAR) imaging quality in the context of a multiaperture acquisition. The analyzed data were recorded over an opportunistic bistatic setup including a stationary ground-based-receiver opportunistic C-band bistatic SAR differential interferometry (COBIS) and Sentinel-1 C-band transmitter. Since the terrain observation by progressive scans (TOPS) mode is operated, the receiver can record synchronization pulses and echoed signals from the scene during many apertures. Hence, it is possible to improve the azimuth resolution by exploiting the multiaperture data. The recorded data are not contiguous and a naive integration of the chopped azimuth phase history would generate undesired grating lobes. The proposed processing scheme exploits the natural sparsity characterizing the illuminated scene. For azimuth profiles recovery greedy, convex, and nonconvex CS solvers are analyzed. The sparsifying basis/dictionary is constructed using the synthetically generated azimuth chirp derived considering Sentinel-1 orbital parameters and COBIS position. The chirped-based CS performance is further put in contrast with a Fourier-based CS method and an autoregressive model for signal reconstruction in terms of scene extent limitations and phase restoration efficiency. Furthermore, the analysis of different receiver-looking scenarios conducted to the insertion in the processing chain of a direct and an inverse Keystone transform for range cell migration (RCM) correction to cope with squinted geometries. We provide an extensive set of simulated and real-world results that prove the proposed workflow is efficient both in improving the azimuth resolution and in mitigating the sidelobes.

Quantum algorithms are designed to process quantum data (quantum bits) in a gate-based quantum computer. They are proven rigorously that they reveal quantum advantages over conventional algorithms when their inputs are certain quantum data or some classical data mapped to quantum data. However, in a practical domain, data are classical in nature, and they are very big in dimension, size, and so on. Hence, there is a challenge to map (embed) classical data to quantum data, and even no quantum advantages of quantum algorithms are demonstrated over conventional ones when one processes the mapped classical data in a gate-based quantum computer. For the practical domain of earth observation (EO), due to the different sensors on remote-sensing platforms, we can map directly some types of EO data to quantum data. In particular, we have polarimetric synthetic aperture radar (PolSAR) images characterized by polarized beams. A polarized state of the polarized beam and a quantum bit are the Doppelganger of a physical state. We map them to each other, and we name this direct mapping anatural embedding, otherwise anartificial embedding. Furthermore, we process ournaturally embeddeddata in a gate-based quantum computer by using a quantum algorithm regardless of its quantum advantages over conventional techniques; namely, we use the QML network as a quantum algorithm to prove that wenaturally embeddedour data in input qubits of a gate-based quantum computer. Therefore, we employed and directly processed PolSAR images in a QML network. Furthermore, we designed and provided a QML network with an additional layer of a neural network, namely, a hybrid quantum-classical network, and demonstrate how to program (via optimization and backpropagation) this hybrid quantum-classical network when employing and processing PolSAR images. In this work, we used a gate-based quantum computer offered by an IBM Quantum and a classical simulator for a gate-based quantum computer. Our contribution is that we provided very specific EO data with anatural embeddingfeature, the Doppelganger of quantum bits, and processed them in a hybrid quantum-classical network. More importantly, in the future, these PolSAR data can be processed by future quantum algorithms and future quantum computing platforms to obtain (or demonstrate) some quantum advantages over conventional techniques for EO problems.

This paper describes an active learning tool for the generation of Earth Observation (EO) benchmark datasets. This tool is able to generate training datasets, based on its active learning strategy with a classification accuracy of around 90%. Afterwards, a data cleaning tool is needed, in order to correct noisy data and provide a clean dataset to be stored in the benchmark database, and for subsequent benchmark verification. The data cleaning procedure is supported by unsupervised learning, using clustering algorithms to group similar patterns, and dimension reduction algorithms to embed them in lower dimension with annotated labels. Moreover, interactive visualizations are implemented in most modules to help better manipulate datasets and get better understandings.

Land cover maps are among the most important products of Remote Sensing (RS) imagery. Despite remarkable advancements in land cover classification techniques, abundant detailed information in the very high-resolution RS images necessitates further improvements to harness the data and discover detailed semantic information. Moreover, scarcity of the labelled data and its quality is a major limitation in RS land cover mapping. In the present study, Latent Dirichlet Allocation is employed for semantic discovery in RS images and a novel kernel-based Bag of Visual Words model is proposed for land cover mapping.

Although the number of labeled datasets in Earth Observation (EO) is increasing, there is still a major gap between the Deep Learning (DL) classifiers designed in this field versus the models in Computer Vision. This gap is produced mainly by the number of datasets available, but also by the diversity of data. In EO, there are different sensors acquiring images, from multispectral (MS) or hyperspectral data, to SAR imagery. In this paper, we want to demonstrate how to reduce the divergence created by the diversity of data. We trained several DL architectures on Bag-of-Words from large-scale MS and SAR datasets, and then we used transfer learning on smaller ones and evaluated the results. With this method, we demonstrate that a DL architecture can be trained with any type of large-scale data, transformed into Bag-of-Words, and the trained model can be used further on other types of data, without regard on the number of channels.

In this work, we address a cross-modal retrieval problem in remote sensing (RS) data. A cross-modal retrieval problem is more challenging than the conventional uni-modal data retrieval frameworks as it requires learning of two completely different data representations to map onto a shared feature space. For this purpose, we chose a photo-sketch RS database. We exploit the data modality comprising more spatial information (sketch) to extract the other modality features (photo) with cross-attention networks. This sketch-attended photo features are more robust and yield better retrieval results. We validate our proposal by performing experiments on the benchmarked Earth on Canvas dataset. We show a boost in the overall performance in comparison to the existing literature. Besides, we also display the Grad-CAM visualizations of the trained model's weights to highlight the framework's efficacy.

Over the last few years, natural disasters elevated dangerously in terms of immensity and prevalence over areas covered by forest and urban woodlands. Fast-spreading nature of the wildfires determine quick uncontrollable situations' causing significant effects in short periods. Despite increased difficulty in image processing approaches due to temporal resolution, complexity of spectral bands and illumination conditions, imagery data streams available from sun-synchronous satellites provide geospatial intelligence in monitoring and preventing fire threats. In this paper, we proposed a local scale burned area estimation framework that employs multispectral images in a deep learning architecture for detecting burned surfaces at patch level. This goal is accomplished by using an autoencoder (AE) network in which the latent feature layer learns normal background distribution, beneficial to background reconstruction. Furthermore, an outlier detection method (OCSVM) is used with aggregated features, latent and covariance components, in order to estimate burned coverage. Our method operates on data retrieved from Sentinel-2 (S2) constellation streaming source, which mainly contain normal scenes and limited fire affected spots.

In this paper, we describe an innovative content annotation method for high-resolution Synthetic Aperture Radar (SAR) images generating routinely user-defined semantic labels for sequences of small contiguous image patches, while the full surface areas of our images cover hundreds of km in width and length. Based on this method, we are able to generate a sea-ice dataset that is used in projects to validate the developed machine learning methods.

This paper aims to present a new algorithm to remove thin clouds and retain information in corrupted images without the use of auxiliary data. By injecting physical properties into the cycle consistent generative adversarial network (GAN), we were able to convert a cloudy multispectral image to a cloudless image. To recover information beneath clouds and shadows we create a synthetic multispectral space to obtain illumination invariant features. Multispectral vectors were transformed from Cartesian coordinates to Polar coordinates to obtain spectral angular distance (SAD) then we employed them as input to train the deep neural network (DNN). Afterward, the outputs of DNN were transformed to Cartesian coordinates to obtain shadow and cloud-free multispectral images. The proposed method, Hybrid GAN-SAD yields trustworthy reconstructed results because of exploiting transparent information from certain multispectral bands to recover uncorrupted images.

In Synthetic Aperture Radar (SAR) image classification tasks, the performance depends on both the classifier and the dataset itself. However, in comparison with plenty of SAR classification methods, there is little work aimed at analyzing the distinguishability of the dataset. In the classification dataset, some classes are semantically different but their distinguishability is low, the classes are hard to be classified especially in some more practical cases that there are unknown classes without supervision exist. Referring to open set recognition (OSR), in this paper, we proposed the SAR Distinguishability Analysor (SAR-DA) to evaluate the distinguishability of the OpenSARUrban dataset. By modeling each class as a multivariate Gaussian distribution in latent space, SAR-DA can not only classify the classes having been seen in training phase, but also can recognize unknown samples if a test sample is out of each known distribution. Each class in OpenSARUr-ban is set unknown in turn, then we apply the SAR-DA on the split dataset in OSR and supervised setting. The distinguishability can be reflected by the unknown recognition recall rate. The experimental results show that the unknown recognition recall rate in OSR setting significantly decreased compared with those in supervised setting, indicating that even though the classes in OpenSARUrban are semantically different from each other, the latent distributions of some classes are quite similar and hard to be classified, thus these classes are of low distinguishability.

Remote sensing multispectral images are extensively used by applications in various fields. The degradation generated by haze or smoke negatively influences the visual analysis of the represented scene. In this paper, a deep neural network based method is proposed to address the visualization improvement of hazy and smoky images. The method is able to entirely exploit the information contained by all spectral bands, especially by the SWIR bands, which are usually not contaminated by haze or smoke. A dimensionality reduction of the spectral signatures or angular signatures is rapidly obtained by using a stacked autoencoders (SAE) trained based on contaminated images only. The latent characteristics obtained by the encoder are mapped to the R - G - B channels for visualization. The haze and smoke removal results of several Sentinel 2 scenes present an increased contrast and show the haze hidden areas from the initial natural color images.

Bistatic radar receivers that use an opportunistic transmitter require a reference channel to capture the original transmitted signal, which is then used as a reference signal for constructing the matched-filter during the range compression step. Because the reference signal is received from line-of-sight, it is orders in magnitude larger than the reflections captured by the receive channel. It is generally difficult to construct the system such that the reference signal is not leaked into the received signal, either via coupling in the circuitry or via reflections off objects in the vicinity of the receiver. Due to its much larger amplitude, the reference signal can easily mask smaller targets with its side-lobes. In this paper we propose a novel deconvolution method for bistatic SAR images as a means of eliminating leakage of the reference signal.

The classification of large-scale high-resolution synthetic aperture radar (SAR) land cover images acquired by satellites is a challenging task, facing several difficulties such as semantic annotation with expertise, changing data characteristics due to varying imaging parameters or regional target area differences, and complex scattering mechanisms being different from optical imaging. Given a large-scale SAR land cover data set collected from TerraSAR-X images with a hierarchical three-level annotation of 150 categories and comprising more than 100 000 patches, three main challenges in automatically interpreting SAR images of highly imbalanced classes, geographic diversity, and label noise are addressed. In this letter, a deep transfer learning method is proposed based on a similarly annotated optical land cover data set (NWPU-RESISC45). Besides, a top-2 smooth loss function with cost-sensitive parameters was introduced to tackle the label noise and imbalanced classes' problems. The proposed method shows high efficiency in transferring information from a similarly annotated remote sensing data set, a robust performance on highly imbalanced classes, and is alleviating the overfitting problem caused by label noise. What is more, the learned deep model has a good generalization for other SAR-specific tasks, such as MSTAR target recognition with a state-of-the-art classification accuracy of 99.46%.

Understanding the physical properties and scattering mechanisms contributes to synthetic aperture radar (SAR) image interpretation. For single-polarized SAR data, however, it is difficult to extract the physical scattering mechanisms due to lack of polarimetric information. Time-frequency analysis (TFA) on complex-valued SAR image provides extra information in frequency perspective beyond the “image” domain. Based on TFA theory, we propose to generate the subband scattering pattern for every object in complex-valued SAR image as the physical property representation, which reveals backscattering variations along slant-range and azimuth directions. In order to discover the inherent patterns and generate a scattering classification map from single-polarized SAR image, an unsupervised hierarchical deep embedding clustering (HDEC) algorithm based on TFA (HDEC-TFA) is proposed to learn the embedded features and cluster centers simultaneously and hierarchically. The polarimetric analysis result for quad-pol SAR images is applied as reference data of physical scattering mechanisms. In order to compare the scattering classification map obtained from single-polarized SAR data with the physical scattering mechanism result from full-polarized SAR, and to explore the relationship and similarity between them in a quantitative way, an information theory based evaluation method is proposed. We take Gaofen-3 quad-polarized SAR data for experiments, and the results and discussions demonstrate that the proposed method is able to learn valuable scattering properties from single-polarization complex-valued SAR data, and to extract some specific targets as well as polarimetric analysis. At last, we give a promising prospect to future applications.

This paper describes the work done with the Data Mining components of the H2020 CANDELA project, mainly the Data Model component on the CANDELA platform as its back end, and the user interaction component of the local user machine as front end. The Data Mining tool is basically composed of four main submodules: the Data Model Generation for Data Mining (DMG-DM), the database management system (DBMS) sub-module that has already been dockerized and deployed on the CANDELA platform, the image search and semantic annotation (KDD) sub-module and the multi-knowledge and query (QE) sub-module. They all require user inputs, and connect directly to the database on the platform; they can be started as a normal GUI (Graphical User Interface) tool.

This paper presents a time-domain synthetic aperture radar (SAR) processor designed for Sentinel-l Terrain Observation by Progressive Scans (TOPS) monostatic data. The processor focuses Interferometric Wide (IW) swath and Extra Wide (EW) swath data on a selected region-of-interest (ROI) and consists of the following main stages: decoding of Level-0 data, selection of the relevant burst and pulses for the targeted ROI, range compression, azimuth frequency unfolding and resampling, and image focusing by a fast subaperture-based version of the back-projection algorithm. The performances of the developed processor are assessed on datasets acquired in IW and EW imaging modes. Such a time-domain processor can be regarded as a first step towards a geometry/frequency-agnostic SAR processing kernel for future monostatic/multi static spaceborne SAR missions.

In this paper, we present three methods that reduce the computational time of training Deep Neural Networks with multispectral images, optimize the resource occupation of the dataset, and obtain high performance for reduced datasets. In the first two methods, we reduce the dimension of the input data with either histograms of pixel intensity or Bag-of-Words. Then we train a Convolutional Neural Network with either histograms or Bag-of-Words and we achieve an accelerated training. Moreover, storing the image patches from the dataset in the form of histograms or Bag-of-Words reduced the memory storage significantly. In the last method, we subsample the training dataset randomly to 50%, 20% and 10% of the original dataset, thus training a Convolutional Neural Network on a smaller number of samples (in the form of histograms or Bag-of-Words), and the classification performance is almost unaffected. This is an important achievement, as there are few labelled datasets for Earth Observation and the number of images in these datasets is small. Our results show that the training time is reduced by a maximum of 387 times and the datasets with histograms or Bag-of-Words occupy 633 times less space.

Due to the complexity and difficult comprehension of remote sensing data, ubiquitous methods for content exploitation are difficult to identify. The main limitation is ascribed to the fact that different perceptions, individual human experts and non-identical remote sensors fabricate a total heterogeneous context. In order to overpass these breaches, an interactive learning model is demanded, where different experts, participating into a validation process, expose and verify shared dictionaries based on users' understanding. We propose a federated system for collaborative labeling of remote sensing datasets that transform blockchain storage into blockchain knowledge. This writing wants to provide insights into a decentralized methodology, capable of building a publicly available, large-scale benchmark data for image classification and hosting a permanently updated machine learning (ML) model.

In this paper is presented a new methodology for synthetic aperture radar images focusing called bidimensional mixed compressive sensing back-projection (CS-BP-2D). Spatial compressibility of the radar images is exploited by constructing the sparsity basis using the backprojection focusing framework and solving the reconstruction problem by means of the orthogonal matching pursuit algorithm (OMP).

The two Sentinel-2 satellites provide, since March 2017, high-resolution worldwide images every five days, freely distributed, generating terabytes of high-dimensional data. An intuitive manner to summarize the main characteristics of the data and gather knowledge is visual exploratory analysis, which is often based on dimensionality reduction methods to represent high-dimensional data. From previous research and the state-of-the-art literature, turned out that t-distributed Stochastic Neighbour Embedding is one of the most appropriate technique to reduce the dimensionality of a dataset, but it requires very high computational power. To overcome this inconvenience, we proposed two hybrid DR algorithms, which combine the DR with the nearest neighbour technique or random forest regression. The main conclusion is that our approaches reduce computational power without compromising the representation quality.

Deep learning based patch-wise Synthetic Aperture Radar (SAR) image classification usually requires a large number of labeled data for training. Aiming at understanding SAR images with very limited annotation and taking full advantage of complex-valued SAR data, this paper proposes a general and practical framework for quad-, dual-, and single-polarized SAR data. In this framework, two important elements are taken into consideration: image representation and physical scattering properties. Firstly, a convolutional neural network is applied for SAR image representation. Based on time-frequency analysis and polarimetric decomposition, the scattering labels are extracted from complex SAR data with unsupervised deep learning. Then, a bag of scattering topics for a patch is obtained via topic modeling. By assuming that the generated scattering topics can be regarded as the abstract attributes of SAR images, we propose a soft constraint between scattering topics and image representations to refine the network. Finally, a classifier for land cover and land use semantic labels can be learned with only a few annotated samples. The framework is hybrid for the combination of deep neural network and explainable approaches. Experiments are conducted on Gaofen-3 complex SAR data and the results demonstrate the effectiveness of our proposed framework.

In this paper, we present our image search system for remote sensing imagery leveraging the capabilities of Elasticsearch, a well-known full-text search engine. We use bag of visual words and bag of visual topics model to represent the earth observation images in a text- equivalent format. The image files are stored in Elasticsearch in their text-equivalent format, making it feasible to apply some sophisticated text-based search functionalities of Elasticsearch on the data. This gives great flexibility to the search clauses allowing the user to match chosen fragments of images. We also implement visual topic search which enables search by higher level semantics. The proposed methods are simple and fast as they do not require any complex image feature computation.

The Markov Random Field (MRF) is used for extracting feature information in images and is formed as an Ising-like model. The quantum annealing is a novel method to optimize objective functions, and objective functions have to be expressed in terms of the Ising model. Hence, the MRF can be embedded into the the quantum annealing method, and feature information of remote sensing images then can be extracted using a quantum annealing computer. Extracted information or features are used to segment an image.

This article presents the achievements of the Candela project. This project aims to develop a platform and new algorithms for the handling, analysis and interpretation of earth observation data. The platform is hosted on the CREODIAS cloud ensuring the proximity of data and its processing. To ensure good performances the platform can scale up or down its computing resources. New algorithms based on machine learning methods for change detection and classification have been developed in the project. The results of these new algorithms are transformed into semantic data used to enrich earth observation products and provide new ways of exploitation. Finally, an end-to-end use of the platform is presented with a use case study of the impact of intense meteorological events on vineyards.

The recent multiplication of open access initiatives to Big Data from Space is giving momentum to the field by widening substantially the spectrum of scientific communities and users, as well as awareness among the public, while offering new benefits at all levels from individual citizens to the whole society. Following a detailed and rigorous review process, 14 articles have been selected out of 48 submissions for this special issue. These are briefly summarized.

Remote sensing (RS) image retrieval is of great significant for geological information mining. Over the past two decades, a large amount of research on this task has been carried out, which mainly focuses on the following three core issues: feature extraction, similarity metric, and relevance feedback. Due to the complexity and multiformity of ground objects in high-resolution remote sensing (HRRS) images, there is still room for improvement in the current retrieval approaches. In this article, we analyze the three core issues of RS image retrieval and provide a comprehensive review on existing methods. Furthermore, for the goal to advance the state-of-the-art in HRRS image retrieval, we focus on the feature extraction issue and delve how to use powerful deep representations to address this task. We conduct systematic investigation on evaluating correlative factors that may affect the performance of deep features. By optimizing each factor, we acquire remarkable retrieval results on publicly available HRRS datasets. Finally, we explain the experimental phenomenon in detail and draw conclusions according to our analysis. Our work can serve as a guiding role for the research of content-based RS image retrieval.

Copernicus is the European programme for monitoring the Earth.It consists of a set of systems that collect data from satellites and in-situ sensors, process this data and provide users with reliable and up-to-date information on a range of environmental and security issues.The data and information processed and disseminated puts Copernicus at the forefront of the big data paradigm, giving rise to all relevant challenges, the so-called 5 Vs: volume, velocity, variety, veracity and value.In this short paper, we discuss the challenges of extracting information and knowledge from huge archives of Copernicus data.We propose to achieve this by scale-out distributed deep learning techniques that run on very big clusters offering virtual machines and GPUs.We also discuss the challenges of achieving scalability in the management of the extreme volumes of information and knowledge extracted from Copernicus data.The envisioned scientific and technical work will be carried out in the context of the H2020 project ExtremeEarth which starts in January 2019.

The paper proposes a methodology to perform azimuth focusing of spaceborne transmitter-stationary receiver bistatic synthetic aperture radar (SAR) data across multiple along-track apertures to increase azimuth resolution. The procedure uses as input several azimuth apertures (continuous groups of range compressed pulses) from one or more satellite bursts and comprises the following stages: azimuth antenna pattern compensation, slow time resampling, reconstruction of missing azimuth samples between neighbouring sets of pulses using an auto-regressive model and back-projection focusing of the resulting multi-aperture range image. The approach is evaluated with real bistatic data acquired over an area of Bucharest city, Romania.

This paper addresses the feasibility of exploiting a radargrammetric procedure for the retrieval of height estimates using stereo images acquired in a space-surface (spaceborne transmitter-stationary receiver) bistatic geometry. Currently, the research interest concerning this particular direction is still in its infancy, as there are very few papers partially covering the subject. The method proposed in this study is applied to a set of SAR images (displaying an urban area of the Bucharest city) in order to assess the elevation of a group of selected targets within the remotely sensed zone.

The increased number of free and open satellite images has led to new applications of these data. Among them is the systematic classification of land cover/use types based on patterns of settlements or agriculture recorded by satellite imagers, in particular, the identification and quantification of temporal changes. In this paper, we will present guidelines and practical examples of how to obtain reliable image patch classification results based on data mining techniques for detecting possible changes that can appear within a data set. Here, we will focus on a scenario, namely forest monitoring using Earth observation Synthetic Aperture Radar data acquired by Sentinel-1, and multispectral data acquired by Sentinel-2.

This paper introduces a tool designed to provide an innovative and insightful way of exploring Earth observation data content beyond visualization, by addressing a visual analytics process. The considered framework combines machine learning and visualization techniques, empowered through human interaction, to gain knowledge from the data. The proposed tool- eVADE leverages the methodologies developed in the fields of information retrieval, data mining and knowledge representation by the means of a visual analytics component. eVADE increases users capability to understand and extract meaningful semantic clusters together with quantitative measurements, presented in a suggestive visual way.

Throughout the years, a wide range of satellite mission enabled the creation of a huge amount of Earth Observation (EO) data carrying complex information, whose exploitation is left behind due to the lack of handling capabilities. Computational resources are hardly keeping up with content analysis and information retrieval. In order to increase the search speed through data warehouses for knowledge discovery, new indexing methods are required to handle both the size and the informational complexity of EO data. Feature extraction algorithms are able to describe the image content, yet, they require a very complex database. In this paper, we propose a methodology that combines feature extraction, hashing methods and optimized indexing to convert the images characteristics into hash codes in an effort to speed up the search process. For our experiments, we run our procedure on a data-set composed of several Sentinel-2 acquisitions form across Europe and we assess the query times.

Deep learning algorithms are widely used in remote sensing image scene understanding. Generally, a large-scale annotated dataset is essential to train a deep neural network for classification. In practical terms, however, a large amount of unknown remote sensing images obtained from different sensors need to be understood which may vary from resolution, geolocation and imaging conditions compared with annotated datasets. In this paper, an unsupervised domain adaptation framework based on ResNet-18 is presented to transfer the knowledge of an existing annotated land cover dataset to other remote sensing data, decreasing the discrepancy among images across sensors. The results show a significant improvement in scene understanding of new remote sensing images.

Sentinel 2 (S2) satellite provides a systematic global coverage of land surfaces, measuring physical properties within 13 spectral intervals at a temporal resolution of 5 days. Computer-based data analysis is highly required to extract similarity by processing and assist human understanding and semantic annotation in support of Earth surface mapping. This paper proposes an exploratory search methodology for S2 data underpinning both visual and latent characteristics by means of data visualization and content representation. For optimized results, the authors focus on a detailed assessment of top relevant state-of-the-art algorithms for features extraction and classification to determine which one could handle best the characteristics of S2 data.

Full-polarimetric synthetic aperture radar (SAR) images have the ability to provide physical patterns of the earth observation, no more than geometric information. In order to learn physical patterns from non-full-polarimetric SAR images, a complex-valued CNN is leveraged to learn a model containing physical parameters. The parameters are learned from the original complex scattering matrix of full-polarimetric SAR images and they can be adopted to extract physical patterns from non-full-polarimetric SAR images. Cloude and Pottier's H-α division, as the annotation principle, is computed by way of coherence matrix. We perform experiments on (German Aerospace Center) DLR's full-polarimetric, airborne F-SAR data, demonstrating that extracting physical patterns from non-full-polarimetric images is feasible. The comparative results illustrate that: 1) The best physical categoric patterns can be extracted from HV and VH polarimetric images in general, while performance from HH and VV polarimetric images are limited; 2) Cross-polarimetric SAR images have greater ability for surface and volume scattering, while co-polarimetric ones are better for multiple scattering extraction.

Single- and dual-polarimetric synthetic aperture radar (SAR) images provide very limited capabilities to interpret physical radar signatures. For generality and simplicity, we call single-polarimetric, dual-polarimetric, and fully polarimetric SAR (PolSAR) images flexible PolSAR images. In order to sufficiently extract physical scattering signatures from this kind of data and explore the potentials of different polarization modes on this task, this paper proposes a contrastive-regulated convolutional neural network (CNN) in the complex domain, attempting to learn a physically interpretable deep learning model directly from the original backscattered data. To achieve a better deep model containing physically interpretable parameters, the objective cost is compared to and selected from several commonly used loss functions in the complex form. The required ground-truth labels are generated automatically according to Cloude and Pottier's H-alpha division plane, which significantly reduces intensive labor cost and transfers this method to an unsupervised learning mechanism. The boundaries between different scattering signatures, however, sometimes show an erroneous separation. With the aim of aggregating intra-class instances and alienating inter-class instances, meanwhile, a complex-valued contrastive regularization term is computed mathematically and is added to the objective cost by a tradeoff factor. Moreover, data augmentation is applied to relieve the side effects caused by data imbalance. Finally, we performed experiments on German Aerospace Center's (DLR)'s L-band, high-resolution (HR), and airborne F-SAR data. Our results demonstrate the possibility of extracting physical scattering signatures from flexible PolSAR images. Physically interpretable potentials of SAR images with different polarization modes are analyzed, and we conclude with physical signature identification.

Object detection is an important and challenging problem in computer vision. Although the past decade has witnessed major advances in object detection in natural scenes, such successes have been slow to aerial imagery, not only because of the huge variation in the scale, orientation and shape of the object instances on the earth's surface, but also due to the scarcity of well-annotated datasets of objects in aerial scenes. To advance object detection research in Earth Vision, also known as Earth Observation and Remote Sensing, we introduce a large-scale Dataset for Object deTection in Aerial images (DOTA). To this end, we collect 2806 aerial images from different sensors and platforms. Each image is of the size about 4000 × 4000 pixels and contains objects exhibiting a wide variety of scales, orientations, and shapes. These DOTA images are then annotated by experts in aerial image interpretation using 15 common object categories. The fully annotated DOTA images contains 188, 282 instances, each of which is labeled by an arbitrary (8 d.o.f.) quadrilateral. To build a baseline for object detection in Earth Vision, we evaluate state-of-the-art object detection algorithms on DOTA. Experiments demonstrate that DOTA well represents real Earth Vision applications and are quite challenging.

Object detection in aerial images plays a significant role in intelligent interpretation of aerial images. Hence many effective methods, especially the new-generation data-driven methods, have been developed for this task. Here, we hold the ODAI, a new contest that focused on object detection in aerial images, based on a new large-scale aerial image dataset called DOTA [1]. This contest contains over 3000 large-size images ( 4k×4k pixels), which cover 211,581 instances divided into 15 categories. Each instance is labeled by an arbitrary (8 d.o.f.) quadrilateral. Besides, we propose two tasks for this contest, named object detection with the horizontal bounding box (OD-HBB) and object detection with the oriented bounding box (OD-OBB). The contest was opened on February 7, 2018, and ended on April 30, 2018. A website is open to the public, which provides links to download data and evaluation server. We have totally received 60 registrations. There are 8 teams that have successfully submitted results on the OD-HBB task with the top mAP as 0.719, and 9 teams that have successfully submitted results on the OD-OBB task with the top mAP as 0.705. Through the contest, we hope to draw extensive attention from a wide range of communities and call for more future research and efforts for the task of object detection in aerial images.

This paper presents the first results obtained by repeat-pass bistatic synthetic aperture radar (SAR) interferometry using a fixed C-band ground-based receiver and the Sentinel-1A/B satellites as transmitters of opportunity. The methodology developed to obtain repeat-pass bistatic SAR interferograms uses as input a stack of range compressed bistatic acquisition data and mainly consists in the following stages: raw inter-ferograms computation on a two-dimensional grid in ground geometry, atmospheric phase screen removal and topographic phase compensation. The displacements of a high-rise building were estimated using two stacks of bistatic SAR images acquired between April-June 2017 over an area of Bucharest city, Romania.

With a continuous increase in multi-temporal synthetic aperture radar (SAR) images, leading to enable mapping applications for Earth environmental observation, the number of algorithms for detection of different types of terrain changes has greatly expanded. In this paper, a SAR image change detection method based on normalized compression distance (NCD) is proposed. The procedure mainly consists in dividing two time series images in patches, computing a collection of similarities corresponding to each pair of patches and generating the change map with a histogram-based threshold. The experimental results were computed using 2 Sentinel 1A images over the city of Bucharest, Romania and 2 TerraSAR-X images over the Elbe River and its surrounding area, Germany.

Synthetic Aperture Radar (SAR) tomography presents the advantage of multiple stable targets detection within same pixel. Fast-sup-GLRT (generalized likelihood ratio test based on support estimation) algorithm proved to be an ideal compromise between detection capabilities and computational complexity. In this work, a multi-look version of this detector which exploits the advantages of Capon estimation is examined. Statistical analysis of estimation and detection processes are conducted to compare the performances of sequential non-linear least-squares (NLLS) search and Capon filtering of projected data for double PS identification. Main objective is to exploit the super-resolution advantages of NLLS method without the risk of multiple stable targets classification from the same scattering contribution. For the last desiderate, an additional verification is included within the detection step.

Advanced interpretation of satellite images calls for automated content analysis as well as interactive content search. A typical example of such systems is EOLib, an ESA funded project that already demonstrated the application potential of TerraSAR-X data within a satellite payload ground segment. In this paper, we analyze the validation results of image content classification using a large set of selected TerraSAR-X images. The classification was done with a cascaded learning method. The main advantage of this method is a coarse-to-fine approach for semantically annotating pixel patches with decreasing size. Once a reliable label is found for a given pixel patch, no further subdivision into still smaller patch sizes is necessary. This leads to a considerable reduction of the computational effort during classification of large-size satellite images.

Past and current EO (Earth Observation) satellite missions have gathered huge amount of data during the past decades. This offers the opportunity to retrieve significant information concerning the evolution of land cover for almost any point of interest (POI) on Earth's surface. This paper presents the evolution of land cover in the administrative area of Bucharest, Romania, over a time span of 33 years. In order to achieve this goal we use data acquired by multiple EO missions such as: Landsat 5 TM (Thematic Mapper), 7 ETM+ (Enhanced Thematic Mapper Plus), 8 OLI/TIRS (Operational Land Im-ager/Thermal InfraRed Sensor) and Sentinel-2 MSI (Multi-Spectral Instrument). We compute several spectral indexes in order to obtain information regarding the surface coverage evolution for categories such as vegetation, water bodies and build up.

Exploratory visual analysis is often required to assist human operator to understand and interpret Earth Observation (EO) images. Optimal image representation offers cognitive support in discovering relevant facts about the scene with respect to a particular application. This is of crucial importance for training data sets selection in all Machine Learning tasks, particularly in the design of active learning tools for multispectral (MS) EO data. This paper proposes a deep neural network (DNN) based method to compress, learn and reveal the most significant information included in the spectral bands of EO data in support of relevant visualization for image content analysis. The advanced method uses a DNN to discover the most suggestive pseudo-color representation able to highlight the entire MS image content better than the particular 3 bands selection (R, G, B). We propose the use of information theory and the concept of mutual information to rank the spectral bands based on the amount of information contained, by applying the minimum-redundancy-maximum-relevance (mRMR) criterion on a the image so that we obtain the ranked bands. A DNN stacked autoencoder based paradigm is developed in order to extract and compress in three bands the overall information from the MS EO data. The developed method is demonstrated and validated for Sentinel 2 dataset.

This paper presents a novel framework for multilabel classification of remote sensing images using Error-Correcting Output Codes (ECOC). Starting with a set of primary class labels, the proposed framework consists in transforming the multiclass problem into binary learning subproblems. The distributed output representations of these binary learners are then transformed into primary class labels. In order to obtain robustness with respect to scale, rotation and image content, a Bag-of-Visual Words (BOVW) model based on Scale Invariant Feature Transform (SIFT) descriptors is used for feature extraction. BOVW assumes an a-priori unsupervised learning of a dictionary of visual words over the training set. Experiments are performed on GeoEye-1 images and the results show the effectiveness of the proposed approach towards multilabel classification, if compared to other methods.

Semantic segmentation or classification for satellite image time series (SITS) is a rarely touched topic, partly due to the difficulty in having the data, but more due to the unreachable task. In this research, we propose a dataset which consists of the Landsat image time series, with the purpose of performing multi-spectral semantic segmentation. As there is no ground truth information, we used unsupervised clustering to group time series into clusters, then Long short term memory (LSTM) unit based Recurrent neural networks (RNN) has been trained. We investigate the accuracy values for our test image patches, around 40% accuracy has been achieved for the sequence classification.

In this letter, a velocity estimation method for moving ships in synthetic aperture radar (SAR) images is proposed based on a subaperture decomposition technique. In contrast to traditional methods, our method needs only a few SAR imaging parameters besides the SAR image itself. The behavior of moving ships in subaperture images is theoretically analyzed, and the ship motion parameters in azimuth direction are accurately estimated. The proposed approach was tested on real SAR stripmap images acquired by TanDEM-X, a twin satellite SAR constellation. The estimated azimuth velocities perfectly fit the data recorded by the international automatic identification system.

Many previous researches have already shown the advantages of multisensor land-cover classification. Here, we propose an innovative land-cover classification approach based on learning a joint latent model of synthetic aperture radar (SAR) and multispectral satellite images using multimodal latent Dirichlet allocation (mmLDA), a probabilistic generative model. It has already been successfully applied to various other problems dealing with multimodal data. For our experiments, we chose overlapping SAR and multispectral images of two regions of interest. The images were tiled into patches and their local primitive features were extracted. Then each image patch is represented by SAR and multispectral bag-of-words (BoW) models. The BoW values are both fed to the mmLDA, resulting in a joint latent data model. A qualitative and quantitative validation of the topics based on ground-truth data demonstrate that the land-cover categories of the regions are correctly classified, outperforming the topics obtained using individual single modality data.

This paper makes an analysis of repeat-pass bistatic synthetic aperture radar (SAR) interferometry performed with a stationary ground-based receiver and a satellite as transmitter of opportunity. A numerical approach is developed in order to asses the sensitivity of the repeat-pass across-track bistatic interferometric phase to height (relative to the digital elevation model used for focusing) and displacements (in the bistatic lines of sight, between consecutive acquisitions). Compared to the monostatic case, the conversion from height/displacement to phase is not straightforward and is dependent on the considered geometry. The method is applied for a bistatic SAR interferogram generated over an area of Bucharest city, Romania, using a ground receiver with one imaging channel and Sentinel-1A/B as transmitter of opportunity.

Synthetic Aperture Radar (SAR) Tomography is a multi-temporal technique which can reconstruct the 3D profile of a scene. One of its main features consists in the ability to detect the presence of multiple Persistent Scatterers (PS) within the same resolution cell. This paper aims to investigate the super-resolution capabilities of SAR Tomography, by detecting targets situated at a distance which is close to Rayleigh resolution. Elevation positions of scatterers are determined with Capon estimation, which is characterized by higher side-lobs reduction. An adapted form of multi-looking is proposed for extraction of targets contribution from reconstructed reflectivity functions and for estimation of reflectivity power textures. An algorithm for detection of dominant and secondary PSs will be conducted based on the derived reflectivity estimates, trying to preserve the advantages of Capon filtering used for reflectivity reconstruction.

An effective monitoring and analysis of ecosystems requires developing new tools and knowledge. In this paper, we propose an approach for detecting land-cover changes using satellite Image Time Series. This approach represents each image by spectral indices and then extracts local features of these representations. Next, a clustering technique (e.g., k-means) is applied to the extracted features, where the resulting clusters are assumed to refer to land-cover classes. The land-cover change is then obtained by counting the number of times an assigned class to each point changes along the time series. For our experiments, we use a collection of Landsat-5 images captured every second month from October 2009 to August 2010 over the protected area of the Doñana National Park in southwestern Spain, which is the largest sanctuary for migratory birds in western Europe. Results demonstrate that the proposed approach can detect the occurring changes in the main land-cover categories along the assessed time series.

This paper presents results obtained by applying the maximum entropy method to image reconstruction of C-band ground-based synthetic aperture radar images. In GB-SARs, azimuth resolution is dependent on the range to target. Hence, a range dependent point spread function is synthesized. Experimental results show that through the maximum entropy method target detection is enhanced resulting in both side lobes reduction and range resolution improvement.

In the quest of developing more accurate methodologies for Earth Observation (EO) image retrieval, visualization and information content exploration, a deep understanding of the data being analyzed is needed. In this paper we propose a simple but efficient visual data mining methodology that can be used for these tasks. Our solution consists in a patch-based feature extraction to derive image features and the projection of the achieved high dimensional feature space in a 3D space using dimensionality reduction methods. Gabor, Spectral Histogram and Bag-of-Words descriptors are the features assigned to represent the content of the data while PCA and t-SNE are the methods designed to achieve the 3D representation. The quality of information provided by the 3D visualization of the data depends on the extracted features. Therefore, a Sentinel-2 scene with various thematic classes is used for feature extraction and classification, to prove the performance of the selected descriptors.

In this work we propose an end-to-end trainable supervised Deep Convolutional Neural Network (DCNN) targeting the task of semantic-segmentation with the addition of class-aware boundary detection. Through this explicit modeling of the class-boundaries, we enforce the network to extract coherent and complete objects, suppressing the uncertainty influencing these regions. Importantly, we show that class-boundary networks in conjunction with DCNN performs optimally, achieving over 90% overall accuracy (OA) on the challenging ISPRS Vaihingen Semantic Segmentation benchmark.

Ground-based fixed receiver bistatic synthetic aperture radar (SAR) is a technology increasingly used in urban monitoring, complementing and enriching the traditional monostatic SAR, but the acquisition geometry is more complex than in the monostatic case. Hence, in the design and configuration of real bistatic SAR systems simulations are needed. In this regard we have presented in [1] a simulator for this geometry that could be used as a way to explore the possibilities given by this configuration. One of the many SAR applications that could be transposed to the bistatic case is displacement measurement. It can be used in urban and non-urban environments, as a way of monitoring the change in the position of some objects of interest, like dams and buildings, using the interferometric phase obtained from two or more SAR acquisitions. This paper aims to investigate by simulation displacement measurements in ground-based fixed receiver bistatic geometry.

Continuously expanding high-resolution and very high resolution multispectral image collections, provided by remote sensing satellites, require specific methods and techniques for data analysis and understanding. Even though there are several patch-based approaches for image classification and indexing, none of them are integrated within a standard. Having the goal to develop an MPEG-7 compliant descriptor for patch-based multispectral earth observation image classification and indexing, we propose a new feature extraction method able to extract maximum information from all the available spectral bands that Sentinel 2, the last generation of remote sensing satellites, provides. Using the polar coordinate transformation of the reflectance values, we obtain illumination invariant features, which can be used along with the scalable color descriptor present in MPEG-7 standard. Also, our method proves to enhance land cover classification of the areas affected by clouds and their shadows and provide similar classification results compared with the homogeneous texture descriptor (HTD), spectral histogram (SH), concatenated HTD with SH features, spectral indices (SIs), and bag-of-words-based descriptors, such as bag-of-SIs and bag-of-spectral-values on cloud-free areas.

We propose a multilevel semantics discovery approach for bridging the semantic gap when mining high-resolution polarimetric synthetic aperture radar (PolSAR) remote sensing images. First, an Entropy/Anisotropy/Alpha-Wishart classifier is employed to discover low-level semantics as classes representing the physical scattering properties of targets (e.g., low-entropy/surface scattering/high anisotropy). Then, the images are tiled into patches and each patch is modeled as a bag-of-words, a histogram of the class labels. Next, latent Dirichlet allocation is applied to discover their higher level semantics as a set of topics. Our results demonstrate that topic semantics are close to human semantics used for basic land-cover types (e.g., grassland). Therefore, using the topic description (bag-of-topics) of PolSAR images leads to a narrower semantic gap in image mining. In addition, a visual exploration of the topic descriptions helps to find semantic relationships, which can be used for defining new semantic categories (e.g., mixed land-cover types) and designing rule-based categorization schemes.

This paper investigates different multi-scale approaches in terms of feature extraction for classification of interfero-metric SAR (InSAR) and previously defined phase gradient InSAR (PGInSAR) images. For this purpose, the scale-space image representation approach is implemented together with the two partial derivative based structure matrices, namely the Hessian matrix and second moment matrix. Their performance is compared to two other multi-scale approaches, namely the Gabor and Fractional Fourier Transform (FrFT) based features, which are quite successful for classification of SAR, InSAR and PGInSAR images. The supervised classification experiments show that the use of PGInSAR images together with the partial derivative based scale-space image representation achieves the best results among all, with a mean accuracy of 90.31% and individual class accuracies more than 80%, even reaching 99% for urban scenes.

This paper presents the first SAR imaging results obtained with a fixed ground-based system used in bistatic configuration having the TerraSAR-X satellite as transmitter of opportunity. The system's characteristics and signal processing flow are presented relative to the state of the art. Compared to previous works on bistatic SAR imaging where a significant amount of processing is dedicated to time/frequency synchronization between the satellite transmitter and ground receiver, we show that a bistatic SAR image with only meter-range geographic offset can be obtained using the state vectors from a monostatic image and minimal synchronization efforts consisting in acquisitions triggered by an amplitude threshold and stabilization of the local oscillator with a GPS-disciplined reference. We present the first bistatic image in ground geometry obtained over an area of Bucharest and compare it with a monostatic image refocused on the same grid.

The main advantage of Synthetic Aperture Radar (SAR) Tomography over classical interferometry consists in capacity to detect the presence of multiple scatterers within the same resolution cell. In this paper we present an algorithm for detection of Persistent Scatterers (PS) based on the variation of reflectivity function in elevation direction reconstructed with Beam-Forming technique. Then, we extract the contribution of dominant scatterers from each resolution cell and reapply the detection algorithm to identify the presence of secondary PSs in scene's pixels. The methods were implemented on a dataset of high-resolution Single Look Complex (SLC) images acquired by TerraSAR-X sensor, on which the residual component of interferometric phase was estimated and compensated locally.

In this paper, we present the visualization of image databases based on their primitive features. Our approach is to have a visual navigation tool for allowing the exploration and exploitation of large image archives. The tool is able to project the content of a given image database based on the primitive feature space and to provide interaction between the final user and the huge amount of data. Land Use/Land Cover area frame statistical Survey in-situ data are used as test dataset. Bag-of-Words and Weber Local Descriptors are used as primitive features.

The data mining systems solve the problem of handling Earth Observation archives counting on a feature vectors based description of the data. Increasing the dimensionality of the feature vectors would offer an effective perspective of the dataset's content. The modern systems provide visual exploration of data projecting their high-dimensional feature space in a 3-D space. The dimensionality reduction methods represent the main way to achieve such representation. Several dimensionality reduction methods have been proposed to identify the mapping, bot not all of them retain the same dataset properties. In order to compare their performance, the development of formal measures like “Trustworthiness” or the measures based on Co-ranking matrix was mandatory. These measures objectively evaluate the similarity between the structure detected in the original and the reduced space. In this paper we evaluate six dimensionality reduction methods using “Trustworthiness” and “Continuity” measures. In this regard three datasets have been used: an artificial one and two remote sensing datasets. Each of them have been described by a high-dimensional feature space.

In this paper we present the early development of a SAR simulator for ground-based fixed-receiver bistatic geometry. Firstly, we describe the assumptions the simulator is based on, then a short presentation of the mathematical model that lays behind the simulator is given. Furthermore, we give some details about its implementation and finally, we present some results.

This paper proposes a custom convolutional deep belief network for polarimetric synthetic aperture radar (PolSAR) data feature extraction. The proposed architecture stands out through the interesting features it shows, starting with the fact that it is adapted to fully polarimetric SAR data. Then, the multilayer approach allows the stepwise discovery of higher-level features. The convolutional approach allows the discovery of local, spatially invariant features and makes the architecture scalable to fully sized PolSAR images. The network is trained in an unsupervised manner, without using labeled data and then it succeeds to extract powerful features from PolSAR patches. This fact is demonstrated by applying supervised and unsupervised classification algorithms on features extracted from patches of a fully polarimetric multi-look F-SAR image over Kaufbeuren airfield, Germany.

The lack of a comprehensive solution for image information mining has often brought confusion and misunderstanding when Earth Observation data based application scenarios were addressed. Considering the variety of dedicated sensors available nowadays, the particularities of the recorded data raises serious issues when explored. Most of the proposed methodologies for data analysis integrate algorithms able to cope with single cases. In order to overcome this limitation, the present paper introduce a compound, configurable framework containing two processing levels, for feature extraction and image classification, that allows different settings depending on the application being handled. The design was proposed such that it facilitates the integration of several methods and algorithm for each level, including a module to serve for validation when reference data is available. The approach is not complete without the interaction with the user, therefore, a human-machine communication strategy was also developed. The validation was performed through a prototype system meeting all the criteria of the defined framework.

Typical applications based on remote sensing imagery start with a visual inspection and selection of scenes and regions of interest to be purchased. In this letter, we propose a so-called Earth observation (EO) epitome as a novel value-added interactive product. Basically, it comprises several image tiles, primitive features, metadata entries, a quick look of the image tiles, and semantic annotations of the image content. It comes with a browser that enables the end user to perform a visual inspection of the image content and semantic annotation of the image objects based on machine learning methods. The epitome offers a more complete description of the content of an EO product. It is envisaged that the epitome will be integrated into the TerraSAR-X Payload Ground Segment and will be offered via user services.

Recently, various patch-based approaches have emerged for high and very high resolution multispectral image classification and indexing. This comes as a consequence of the most important particularity of multispectral data: objects are represented using several spectral bands that equally influence the classification process. In this letter, by using a patch-based approach, we are aiming at extracting descriptors that capture both spectral information and structural information. Using both the raw texture data and the high spectral resolution provided by the latest sensors, we propose enhanced image descriptors based on Gabor, spectral histograms, spectral indices, and bag-of-words framework. This approach leads to a scene classification that outperforms the results obtained when employing the initial image features. Experimental results on a WorldView-2 scene and also on a test collection of tiles created using Sentinel 2 data are presented. A detailed assessment of speed and precision was provided in comparison with state-of-the-art techniques. The broad applicability is guaranteed as the performances obtained for the two selected data sets are comparable, facilitating the exploration of previous and newly lunched satellite missions.

Modern knowledge discovery systems, empowered by visual data exploration techniques, enable the user to discover and understand the data content. Considering patch-level processing, the visual exploration of Earth Observation archives aims to identify groups of items sharing similar semantic content. Each patch is further represented by certain descriptors, i.e., spectral signatures or Weber local descriptors, to capture structural signature. Later on, the content of the archive is illustrated by a 3-D projection of the high-dimensional space of the descriptors. Aspiring to prove the visual data mining potential, this letter intends to determine the capability of dimensionality reduction techniques to achieve a meaningful 3-D projection of the high-dimensional space. Several real-world data sets were used, i.e., University of California, Merced Land Use data set and a Landsat 7 Enhanced Thematic Mapper Plus image tiled into patches.

Deep learning methods such as convolutional neural networks (CNNs) can deliver highly accurate classification results when provided with large enough data sets and respective labels. However, using CNNs along with limited labeled data can be problematic, as this leads to extensive overfitting. In this letter, we propose a novel method by considering a pretrained CNN designed for tackling an entirely different classification problem, namely, the ImageNet challenge, and exploit it to extract an initial set of representations. The derived representations are then transferred into a supervised CNN classifier, along with their class labels, effectively training the system. Through this two-stage framework, we successfully deal with the limited-data problem in an end-to-end processing scheme. Comparative results over the UC Merced Land Use benchmark prove that our method significantly outperforms the previously best stated results, improving the overall accuracy from 83.1% up to 92.4%. Apart from statistical improvements, our method introduces a novel feature fusion algorithm that effectively tackles the large data dimensionality by using a simple and computationally efficient approach.

In this paper, the importance of image transformation for parametric modeling of single-look complex (SLC) and in-terferometric SAR (InSAR) images is emphasized. For SLC images, the real and imaginary parts of the fractional Fourier transform (FrFT) coefficients have already been modeled with generalized Gaussian distribution (GGD). Here, this work is extended for InSAR images. The Kolmogorov-Smirnov (KS) test statistics show that FrFT simplifies the statistical response for both SLC and InSAR images, and helps to achieve more uniform KS statistics over all classes, which is important in order to model the whole database with a single distribution. Moreover, the classification of InSAR images with a feature vector composed of GGD parameters shows a performance comparable to that of a non-parametric feature vector.

In this paper, we consider the problem of satellite image classification, in which feature extraction is a critical step. One of the most prevalent methods is the Bag-of-Words (BoW) feature representation, which attains state-of-the-art performance in many applications. It has five steps: feature detection, local feature extraction, dictionary learning, feature coding, and feature pooling. In this paper, we focus on the second and third step. We propose a simple yet efficient feature extraction method within the BoW framework. It has two main advantages. Firstly, this method does not need any complex local feature extraction; instead, it uses directly the pixel values from small windows as low level features. Secondly, instead of using a time-consuming clustering algorithm for dictionary learning, a random dictionary is built and applied to feature space quantization. An extensive experimental evaluation has been performed and compared with other feature extraction methods. It is demonstrated that our feature extraction method is quite competitive for optical and SAR satellite image classification.

Simple color, intensity representations of polarimetric synthetic aperture radar (PolSAR) images fail to show the physical characteristics of the recorded ground objects, so several coherent and incoherent decomposition theorems have been proposed in the state-of-the-art literature. All these decompositions assume the fact that any scattering mechanism can be represented as the sum of some simpler, canonical" scattering mechanisms. Following the same assumption, in this paper we employ the independent component analysis (ICA) for PolSAR images representation. Since ICA is a method used for blind sources separation, we expect that the derived ICA channels represent as well as possible certain types of scattering mechanisms present in the image. ICA decomposition is validated against the coherent Pauli and the incoherent H/a/α decompositions. The normalized compression distance (NCD) is used as a measure of quality of decompositions. Experiments are made on a SLC L-band F-SAR image over Kaufbeuren airfield, Germany.

In this paper we present the LUCAS Visual Browser system, a tool for land cover visual analytics. The system implements different web technologies in a multilayer server-client architecture in order to allow the user to visually analyse land cover heterogeneous information. The information manage is composed of EO multispectral and SAR products along with the multitemporal in situ LUCAS surveys. The fusion of these data provides a very useful information during the EO scene interpretation process. Furthermore, the system offers interactive tools for the detection of optimal datasets for EO multi-temporal image change detection, providing at the same time ground truth points for both, human and machine analysis.

Feature learning algorithms aim to provide a compact and discriminative representation of complex datasets in order to increase the speed and accuracy of clustering or classification. In this paper, we propose a novel interactive feature learning approach which is mainly based on 3D interactive data visualization and Non-negative Matrix Factorization (NMF). Here, the data is visualized in a 3D interface to support human-data interaction. The user interactions are exploited in an NMF framework to learn a compact representation of the data. The conducted experiments on Synthetic Aperture Radar (SAR) images confirm the efficiency of the proposed approach.

In many applications, such as image retrieval and change detection, we need to assess the similarity of two statistical models. As a distance measure between two probability density functions, Kullback-Leibler divergence is widely used for comparing two statistical models. Unfortunately, for some models such as Gaussian Mixture Model (GMM), Kullback-Leibler divergence has no analytically tractable formula. We have to resort to approximation methods. In this paper, we compare seven methods, namely Monte Carlo method, matched bond approximation, product of Gaussian, variation-al method, unscented transformation, Gaussian approximation, and min-Gaussian approximation, for approximating the Kullback-Leibler divergence between two Gaussian mixture models for satellite image retrieval. Two image retrieval experiments based on two publicly available datasets have been performed. The comparison is carried out in terms of both retrieval performance and computational time.

Recently, two improved methods have shown their advantages in browsing Earth Observation (EO) dataset. The first method is the Bag-of-Words (BoW) feature extraction method and the second is the Normalized Compression Distance (NCD) for assessing image similarity. However, they have not been compared so far for satellite image retrieval, which motivates this paper. Two retrieval experiments have been performed on a freely available optical image dataset and a SAR image dataset. Through these two experiments, we conclude that the BoW method performs generally better than NCD. Although it is a parameter-free solution for data mining, NCD only performs well for images with repetitive patterns like some homogeneous classes. In contrast, BoW method performs much far beyond that of NCD. In addition, NCD is computationally very expensive, which makes it infeasible to be applied in real applications. In contrast, BoW method is more realistic in practical applications in terms of both accuracy and computation.

Visual data mining methods are of great importance in exploratory data analysis having a high potential for mining large databases. As the data feature space is generally n-dimensional, visual data mining relies on dimensionality reduction techniques. This is the case for image feature spaces which can be visualized by giving each data point a location in a three dimensional space. This paper aims to present a comparative study of several dimensionality reduction methods considering as input image feature spaces, in order to detemine an optimal visualization method to illustrate the separation of the classes. At the beginning, to check the performance of the envisaged method, an artificial dataset consisting of random vectors describing six, 20-dimensional Gaussian distributions with spaced means and low variances was generated. Further, two real images datasets are used to evaluate the contributions of dimensionality reduction algorithms related to data visualization. The analysis focuses on the PCA, LDA and t-SNE dimensionality reduction techniques. Our tests are performed on images for which the computed features include the color histogram and Weber descriptors.

As the data acquisition capabilities of Earth observation (EO) satellites have been improved substantially in the past few years, large amount of high-resolution satellite images are downlinked continuously to ground stations. Such amount of data increases rapidly beyond the users' capability to access the images' content in reasonable time. Hence, automatic and fast interpretation of a large data volume is a computationally intensive task. Recently, approximate nearest neighbour search has been used for content-based image retrieval in sub-linear time. Kernelized locality sensitive hashing (KLSH) is a well-known approximate method, which has recently shown promising results for fast remote sensing image retrieval. This paper proposes a novel parallelization of KLSH using Graphical Processing Units (GPU), in order to perform fast parallel image retrieval. The proposed method was tested on high-dimensional feature vectors from two satellite-based image datasets, where an average speedup of 20 times was achieved.

Satellite image time series are a valuable resource for enhancing land exploitation by respecting the natural cycles, analyzing urban expansion and its positive and negative effects, limiting the unhealthy rhythm of deforestation, understanding natural hazards and so on. In this context, understanding only the changes in multitemporal images is not sufficient. This paper aims to correlate multi-level change detection techniques with image semantic segmentation methods in order to build an hierarchy of changes for each semantic class. In this way, we are able to provide statistics regarding the levels of change suffered by a certain area. The methods are demonstrated with examples involving bi-temporal Land-sat images.

This paper makes a comparative assessment of the observable landcover classes visible in the data provided by the newly launched SENTINEL-1 (S-1) satellite. The analysis focuses on two feature extraction methods previously reported in the literature to be able to distinguish between a relatively large number of classes in high and very high resolution SAR data. The analysis of the S-1 medium resolution data makes use of the Gabor filtering and Fourier spectral coefficients. Moreover, we consider the opportunity of speckle reduction before feature extraction, considering that the texture analysis is sensitive to correlated speckle. Furthermore, the comparison takes into account the adaptation of the window size to the resolution and pixel spacing of the data. In order to make a quantitative assessment of the results, we perform a joint evaluation of the detection probability at pixel and patch level, with respect to an expert annotated dataset.

Several algorithms for polarimetric synthetic aperture radar (PolSAR) data indexing and classification were proposed in the state of the art literature. In particular, one of them computes powerful, compact feature descriptors composed of the first three logarithmic cumulants of the BiQuaternion Fractional Fourier Transform (BiQFrFT) coefficients of PolSAR patches. Since the BiQFrFT of each patch is computed at three different angles, the algorithm's result consists in nine complex-valued features (18 real-valued features) for single polarization images and in nine biquaternion-valued features (72 real-valued features) for fully polarimetric images. In this paper feature selection based on mutual information (MI) is employed to optimally select a subset of features, in order to improve the indexing performances and to minimize the classification error. The improved results are shown on two polarimetric images: a L-band PALSAR image over Danube's Delta, Romania and a C-band RadarSAT2 image over Brâila, Romania.

In this paper, we present a Gaussian test-based hierarchical clustering method for high-resolution TerraSAR-X images. The purpose is to obtain homogeneous clusters. k-means is used to split image features to create a hierarchical structure. As image feature vectors usually fall into high dimensional feature space, we test different distance metrics, in order to try to tackle the curse of dimensionality problem. With prepared datasets, we evaluate the clustering results by defining a homogeneity percentage. The results show that by using Gabor texture feature, the Gaussian test-based hierarchical patch clustering method is able to obtain homogeneous clusters. Meanwhile, fractional distance or Minkowski distance performs better than Euclidean or Manhatten distance.

The large volume of detailed land cover features, provided by high resolution Earth observation (EO) images, has attracted considerable interest in the discovery of these features by learning systems. In this letter, we perform latent Dirichlet allocation on the bag of words (BoW) representation of collections of EO image patches to discover their semantic-level features, the so-called topics. To assess the discovered topics, the images are represented based on the occurrence of different topics, called bag of topics (BoT). The value added by BoT to the BoW model of image patches is then measured based on existing human annotations of the data. In our experiments, we compare the classification accuracy results of BoT and BoW representations of two different remote sensing image data sets, a multispectral optical data set and a synthetic-aperture-radar data set. Experimental results demonstrate that BoT can provide a compact and semantically meaningful representation of data; it either causes no significant reduction in the classification accuracy or increases the accuracy by a sufficient number of topics.

Research on the semantic gap has considered differences between user and computer image interpretations and proposed methods to bridge it. These methods have been verified by comparing results to reference data or by measuring the degree of user acceptance. Although these methods result in a narrower semantic gap between computers and users, the resulting model for a specific user and search goal may still not be satisfactory to other users. Through an image annotation task with users, we find that this discrepancy is caused by the subjective biases present in the bridging methods, which we refer to as the “linguistic semantic gap.” Based on our findings, efforts to bridge the semantic gap should include different user perspectives to compensate the individual subjective biases, by increasing the diversity of data sets used in the domain. Moreover, models derived from proposed bridging methods could be stored and further used by other systems.

In the literature, scene recognition from interferometric synthetic aperture radar (InSAR) images has been mainly focused on the joint use of the backscatter intensity and the coherence between interferometric image pairs. However, the terrain height information residing in the interferometric phase requires further exploration for classification purposes. In this letter, taking the interferometric phase information into account together with the backscatter intensity, the whole complex- valued InSAR image is exploited for feature extraction. In addition, a new complex-valued phase-gradient InSAR (PGInSAR) image is defined. A fractional-Fourier-transform-based feature ex traction, which was proposed for the classification of single-look complex (SLC) SAR images, is adopted for InSAR and PGInSAR images. For patch-based classification, an image database is generated from bistatic pairs acquired from the same terrain with three different effective baselines. The supervised κ-nearest neighbor classification results show that InSAR outperforms SLC by 15%, whereas PGInSAR introduces further 10% improvement over InSAR or a total improvement of 27% over SLC. Moreover, PGInSAR is found to be more robust to effective baseline changes than InSAR, which makes PGInSAR a better candidate for feature extraction.

In this letter, we present a novel technique for unsupervised change analysis that leads to a method of ranking the changes that occur between two satellite images acquired at different moments of time. The proposed change analysis is based on binary descriptors and uses the Hamming distance as a similarity metric. In order to render a completely unsupervised solution, the obtained distances are further classified using vector quantization methods (i.e., Lloyd's algorithm for optimal quantization). The ultimate goal in the change analysis chain is to build change intensity maps that provide an overview of the severeness of changes in the area under analysis. In addition, the proposed analysis technique can be easily adapted for change detection by selecting only two levels for quantization. This discriminative method (i.e., between changed/unchanged zones) is compared with other previously developed techniques that use principal component analysis or Bayes theory as starting points for their analysis. The experiments are carried on Landsat images at a 30-m spatial resolution, covering an area of approximately 59×51 km2over the surroundings of Bucharest, Romania, and containing multispectral information.

Dramatic growth in the volume of data made a compact and informative representation of the data highly demanded in computer vision, information retrieval, and pattern recognition. Non-negative Matrix Factorization (NMF) is used widely to provide parts-based representations by factorizing the data matrix into non-negative matrix factors. Since non-negativity constraint is not sufficient to achieve robust results, variants of NMF have been introduced to exploit the geometry of the data space. While these variants considered the local invariance based on the manifold assumption, we propose Farness preserving Non-negative Matrix Factorization (FNMF) to exploits the geometry of the data space by considering non-local invariance which is applicable to any data structure. FNMF adds a new constraint to enforce the far points (i.e., non-neighbors) in original space to stay far in the new space. Experiments on different kinds of data (e.g., Multimedia, Earth Observation) demonstrate that FNMF outperforms the other variants of NMF.

Automated recognition of SAR images requires feature extraction from complex-valued data. In this work, complex-valued interferometric SAR (InSAR) images, which are mainly used to construct elevation models, are proposed for feature extraction. Feature extraction based on the log-cumulants of fractional Fourier transform (FrFT) coefficients has already been proposed in the literature and found to be quite successful for the classification of single-look complex (SLC) SAR images. Here, this method is applied to the complex-valued InSAR and newly introduced complex-valued phase gradient InSAR (PGInSAR) images. In order to evaluate the classification performance for SLC, InSAR and PGInSAR images, a database of bistatic TanDEM-X interferometric pairs is constructed, and a supervised KNN classification is performed. The overall classification accuracies for a 4% training set size show that the use of InSAR and PGInSAR images outperforms SLC images by 15% and 24%, respectively. In terms of individual class accuracies, the biggest improvement is observed in agricultural fields and mixed vegetation areas.

In this paper we present the result of data analytics techniques applied to a database comprising of 32 SLC SM TerraSAR-X images, acquired over the area of Bucharest, Romania. The methodology follows a two step approach. The first stage consists of a coarse identification of potentially changed areas using a supervised learning image annotation tool with relevance feedback. Gabor texture features are used to describe image patches. The patch size is derived as a function of the resolution and pixel spacing of the data. In the second stage we apply an information theory strategy to refine the regions previously shown to exhibit class dynamics within the image stack, with pixel accuracy. Finally, a series of analytical indicators (absolute extent of areas affected by change, class evolution trends, inter-class correlations) are derived, in order to generate a predictive model for the selected test site.

In this paper, we present an accelerated knowledge-driven content-based information mining system for Big Earth Observation data fusion. The tool combines, at pixel level, the unsupervised clustering results of different number of features. The features, extracted from different EO raster image types and from existing GIS vector maps, are combined, in form of a BoW, with a user given semantic concepts in order to calculate the posterior probability that allows the final search. The inclusion of GIS data during the active learning, based on Bayesian networks, accelerate the definition processes of semantic labels and retrieve the related images with only a few user interactions. The inclusion of GIS data in conjunction with the recently introduced search algorithm have as a result a system which greatly optimizes the computational costs and over performs existing similar systems in various orders of magnitude.

In this paper, we present a quantitative analysis for a rapid mapping scenario that performs a damage assessment of the 2013 floods in Germany. The scenario is created using pre-disaster and post-disaster TerraSAR-X images and an automated annotation system. Our data set is tiled into patches and Gabor filters are used as a primitive feature method applied to each patch separately. An active learning system based on support vector machine is implemented in order to group the features into categories. Once all categories are identified, these are semantically annotated using reference data as ground truth. In our evaluation 7 categories were retrieved with their specific taxonomies defined using our previous hierarchical annotation scheme. We show that the system supports rapid mapping scenarios (e.g., floods, tsunami, earthquake, etc.) and interactive mapping generation. In addition, with the help of this system, quantitative assessment of disasters can be carried out.

The latest generation of synthetic aperture radar (SAR) instruments operating in X-band, that is, COSMO-SkyMed (CSK) and TerraSAR-X (TSX), are capable of providing images from coarse resolution to very high resolution. A lot of research effort has been invested in the study and understanding of images obtained from these satellites. However, there is still a huge scope of statistical understanding and comparison of data from both satellites. In this study, we demonstrate some striking similarities between medium resolution data obtained from CSK and TSX Stripmap mode images. Landcover unsupervised clustering using k-means is discussed to further justify our findings. Clustering is carried out using a feature descriptor based on log-cumulants of Gabor coefficients, which was recently proposed by us in earlier studies.

The amount of data available on the internet provides massive additional information for the Earth Observation (EO) imagery. Periodical news, various reports and measurements, pictures or online encyclopedias are just few examples of the existent information. Occasionally, this data offers new perspectives for EO image understanding and interpretation. However, current image analysis do not benefit from the advantage given by external sources. To overcome these drawbacks, the present paper proposes an approach that goes beyond traditional information mining by using a joint image and text analysis. Fast Compression Distance (FCD) is computed to measure the similarities inside a collection of very high resolution images and text files. The main purpose is to discover common patterns within the data, without any a priori assumption, parameter-free, relying on data compression-based techniques. A hierarchical clustering is performed in order to learn about the dependencies between different types of data.

In this letter, we carry out a comparative study of statistical models for multilook synthetic aperture radar amplitude images. Ten state-of-the-art statistical models are selected for comparison. To achieve a fair evaluation, we estimate all model parameters using the method of log-cumulants and apply the method to an image pyramid with varying pixel spacing (and resolution). The pyramid is created by different image product generation options. In addition to pixel spacing and resolution, we also consider the homogeneity of a scene for performance evaluation and we apply three performance measures. Through this study, it was found out that some models perform well for all resolutions, while the performance of other models depends heavily on the image content.

This paper presents an information extraction and image enhancement technique using single-look complex (SLC) synthetic aperture radar data. The novelty of this method is the proposed complex-domain despeckling stage. Tikhonov-like optimization is used for minimizing the cost function, which consists of a Gauss-Markov random field (GMRF) prior. The GMRF model is used for texture modeling. The texture parameters of the GMRF are estimated using the evidence maximization framework. The experimental results showed that despeckled SLC images have well-preserved textural features, structures, and point scatterers. The phase of the reconstructed image is well preserved and provides good-quality interferograms of high-resolution spotlight images.

We are dealing with large-scale high-dimensional image data sets requiring new approaches for data mining where visualization plays the main role. Dimension reduction (DR) techniques are widely used to visualize high-dimensional data. However, the information loss due to reducing the number of dimensions is the drawback of DRs. In this paper, we introduce a novel metric to assess the quality of DRs in terms of preserving the structure of data. We model the dimensionality reduction process as a communication channel model transferring data points from a high-dimensional space (input) to a lower one (output). In this model, a co-ranking matrix measures the degree of similarity between the input and the output. Mutual information (MI) and entropy defined over the co-ranking matrix measure the quality of the applied DR technique. We validate our method by reducing the dimension of SIFT and Weber descriptors extracted from Earth Observation (EO) optical images. In our experiments, Laplacian Eigenmaps (LE) and Stochastic Neighbor Embedding (SNE) act as DR techniques. The experimental results demonstrate that the DR technique with the largest MI and entropy preserves the structure of data better than the others.

The collected Earth Observation (EO) data volumes are increasing immensely. In the meantime, the need for retrieval of focused information for decision making is increasing. Due to the particular nature of EO sensors, recording signals very differently than humans perceptual system, the challenges raised by the semantic and sensory gaps are immensely amplified in designing retrieval methods for EO images. This article introduces a method based on communication channel model to quantify and measure the semantic gap, used to assess various feature descriptors for semantic annotation purposes. The approach uses Latent Dirichlet Allocation (LDA), considering images as the source and the semantic topics as the receiver. The parameters of LDA are estimated for computing the Mutual Information to assess latent semantics of feature space. We further introduce a method to measure the distance between humans' and computer's semantics. The results are validated using an SVM-based classifier for an annotated dataset.

In high-resolution (HR) and very-high resolution (VHR) synthetic aperture radar (SAR) images, focus is now on the patch-oriented image categorization in contrast to the pixel-based classification in low-resolution SAR images. SAR image categorization requires the generation of a compact feature descriptor that accurately defines the content of the image patch under consideration. In this paper we propose a parametric feature descriptor generated on the complex-valued SAR image within a transformation space. The fractional Fourier transform (FrFT), has been considered to transform the image pixels of the single-look complex (SLC) SAR images in order to obtain a simpler statistical response. The real and imaginary components of the complex-valued FrFT coefficients have been modelled with generalized Gaussian distribution (GGD). The proposed feature descriptor is compared with a Wavelet-decomposition-based parametric feature descriptor; and with the FrFT-based and Gabor-filter-bank-based non-parametric feature descriptors. Categorization accuracy enhancement is demonstrated over several categories comprising of natural topologies. The experimental database consists of 2000 image patches (of size 200 × 200 pixels) extracted from SLC HR TerraSAR-X scenes.

In this paper, we propose to identify the number of categories that can be retrieved from a very high resolution SAR data. The evaluation is done on TerraSAR-X high resolution Spotlight data and the retrieved categories are semi-automatically annotated using as feature vector the Gabor filters; as a classifier the Support Vector Machine, and for ranking the suggested images the relevance feedback. The visualization of the tool was enhanced compared with our previous implementation in order to support the users in his/her approach to search the patches of interest in a large repository. Our dataset consist in 43 scenes that cover as much as possible all the regions over the world. A total of 352 categories are identified that contain urban and non-urban categories.

The increasing number of Earth Observation image acquisitions provides huge volume of information, which requires new techniques, methods and tools to explore and exploit the abundant volume of information contained in the image archives. This paper presents an architecture concept for data mining systems using Earth Observation images. The architecture concept is defined as a modular system composed of five modules allowing functions such as primitive feature extraction, image content and context analyses, finding scenes of interest by content, enriched metadata, and semantics, the interpretation and understanding of the image content, semantic definition of the image content, and visualization of the image database via human machine interfaces. All these functionalities are integrated and supported by a database management system.

Recent development in the design of modern satellite ground segments include systems and tools for automated content analysis allowing users to conduct systematic semantic searches within satellite image data archives. The need for such tools becomes more and more pressing as future space-borne imaging sensors will deliver enormous quantities of data that cannot be studied manually. For instance, typical examples from a European perspective are described in [1] and [2]. Within this framework, the European Space Agency (ESA) has started to fund the Earth Observation Librarian (EOLib) project to set up the next generation of image information mining systems [3]. Here we report on the preparation of scenarios that are needed for training and to verify and optimize the performance of such systems.

The TerraSAR-X satellite has been fully operational since nearly 6 years and has delivered a very large quantity of high resolution SAR images. Among these images one can find a number of repeated acquisitions of selected target areas taken with nearly identical imaging parameters. These image time series data often cover the full seasonal cycle of a target area and lend themselves well to small-scale change detection; however, in the case of urban areas, we have to be aware of the quantitative impact of rain, frost and wind on high resolution SAR images. In particular, images of Western European cities are characterized by construction work concentrating on single buildings within a fully built-up city and by public green space changes. A quantitative analysis of urban time series data has to discriminate between definite changes of the urban landscape and the transient impacts of rain, frost and wind. The critical issue is how to identify and characterize these transient phenomena.

Efficient management and exploration of high-volume scientific file repositories have become pivotal for advancement in science. We propose to demonstrate the Data Vault, an extension of the database system architecture that transparently opens scientific file repositories for efficient in-database processing and exploration.

With the advent of very high resolution (VHR) synthetic aperture radar (SAR) images, local content description is becoming a critical issue for indexing. Conventional SAR image analysis techniques, like segmentation and pixel-level classification, are likely to fail as high-level semantic description should be considered for better discrimination. Therefore, we propose to use image-patch-based analysis method for SAR image interpretation. Inspired by ratio edge detector, in this letter, a new feature extraction method represented by the mean ratios in different directions is proposed for VHR SAR image content characterization. Based on the mean ratio, two simple yet powerful and robust features are proposed for SAR image patch indexing. One is the bag-of-word model using not only the basic statistics, i.e., local mean and variance, but also the mean ratios in different directions. The second one is an adaptation of the Weber local descriptor to SAR images by substituting the gradient with the ratio of mean differences in vertical and horizontal directions. To evaluate the proposed features, image patch indexing based on active learning using a SAR image database consisting of high-resolution TerraSAR-X patches is performed. Comparison with the state-of-the-art features, particularly texture features, has shown improved performance for SAR image categorization.

In this paper, we propose to study the dependence of information extraction technique performance on synthetic aperture radar (SAR) imaging parameters and the selected primitive features (PFs). The evaluation is done on TerraSAR-X data, and the interpretation is realized automatically. In the first part of this paper (use case I), the following issues are analyzed: 1) finding the optimal TerraSAR-X products and their limits of variability and 2) retrieving the number of categories/classes that can be extracted from the TerraSAR-X images using the PFs (gray-level co-occurrence matrix, Gabor filters, quadrature mirror filters, and nonlinear short-time Fourier transform). In the second part of this paper (use case II), we investigate the invariance of the products with the orbit direction and incidence angle. On the one hand, the results show that using ascending looking is better than using descending looking with an average accuracy increase of 7%-8%, approximately. On the other hand, the classification accuracy for the incidence angle varies from a lower value of the incidence to an upper value of the incidence angle (depending on the sensor range) with 4%-5%. The test sites are Venice (Italy), Toulouse (France), Berlin (Germany), and Ottawa (Canada) and are covering as much as possible the huge diversity of modes, types, and geometric resolution configuration of the TerraSAR-X. For the evaluation of all these parameters (resolution, features, orbit looking, and incidence angle), the support-vector-machine classifier is considered. To evaluate the accuracy of the classification, the precision/recall metric is calculated. The first contribution of this paper is the evaluation of different PFs (proposed in the literature for different types of images) and adaptation of these for SAR images. These features are compared (based on the accuracy of the classification) for the first time for a multiresolution pyramid specially built for this purpose. During the evaluation, all the classes were annotated, and a semantic meaning was defined for each class. The second main contribution of this paper is the evaluation of the dependence on the patch size, orbit direction, and incidence angle of the TerraSAR-X. This type of evaluation has not been systematically investigated so far. For the evaluation of the optimal patch, two different patch sizes were defined, with the constrained that the size on ground needs to cover a minimum of one object (e.g., 200 × 200 m on ground). This patch size depends also on the parameters of the data such as resolution and pixel spacing. The investigation of orbit looking and incidence angle is very important for indexing large data sets that has a higher variability of these two parameters. These parameters influence the accuracy of the classification (e.g., if the incidence angle is closer to the lower bounds or closer to the upper bound of the satellite sensor range).

Advances in the image retrieval (IR) field have contributed to the elaboration of tools for interactive exploration and extraction of the images from huge archives associating the content of the images with semantic meaning. This paper presents an Earth-observation (EO) IR system based on enriched metadata, semantic annotations, and image content called EO retrieval. EO retrieval generates an EO-data model by using automatic feature extraction, processing the EO product metadata, and defining semantics, which later is fully exploited for supporting complex queries. In order to demonstrate the functionality of the system, we have created a semantic catalog of TerraSAR-X as application scenario. The database is composed of 39 high-resolution TerraSAR-X scenes comprising about 50 000 image patches (160 × 160 pixels) with their feature descriptors, 100 of metadata entries for each scene, and about 330 semantic annotations. Many query examples combining semantics, metadata, and image content for full exploitation of the image database are presented.

With the advent of high-resolution (HR) synthetic aperture radar (SAR) images from satellites like TerraSAR-X and TanDEM-X, interest is now on patch-oriented image categorization in contrast to the pixel-based classification in low-resolution SAR images. SAR image categorization requires the generation of a compact feature descriptor that accurately defines the content of the image patch under consideration. As phase information plays a critical role in SAR images, this paper proposes the use of a chirplet-derived transform, i.e., the fractional Fourier transform (FrFT), for generating a compact feature descriptor for single-look complex (SLC) SAR images. Representing a SAR signal in rotated joint time-frequency planes via the FrFT allows discovering the underlying backscattering phenomenon of the objects on the ground. SAR image projections on different planes of the joint time-frequency space using the FrFT provide a simple statistical response that is easier to analyze. The proposed method has been compared with a multiscale approach, i.e., Gabor filter banks, a second-order-statistics-based method (as gray-level co-occurrence matrices), and a spectral descriptor method. We demonstrate the suitability of the FrFT-based method for image categorization on the basis of backscattering behavior, whereas the Gabor-filter-bank-based method is found mainly suitable for images with a strong texture. This paper demonstrates enhancement in the separability for most of the considered categories when using logarithmic cumulants instead of linear moments for both the FrFT-based and Gabor-filter-bank-based methods. The experimental database consists of 2000 image patches (of size 200 × 200 pixels) extracted from SLC HR TerraSAR-X scenes.

With a continuous increase in the number of Earth Observation satellites, leading to the development of satellite image time series (SITS), the number of algorithms for land cover analysis and monitoring has greatly expanded. This paper offers a new perspective in dynamic classification for SITS. Four similarity measures (correlation coefficient, Kullback-Leibler divergence, conditional information, and normalized compression distance) based on consecutive image pairs from the data are employed. These measures employ linear dependences, statistical measures, and spatial relationships to compute radiometric, spectral, and texture changes that offer a description for the multitemporal behavior of the SITS. During this process, the original SITS is converted to a change map time series (CMTS), which removes the static information from the data set. The CMTS is analyzed using a latent Dirichlet allocation (LDA) model capable of discovering classes with semantic meaning based on the latent information hidden in the scene. This statistical method was originally used for text classification, thus requiring a word, document, corpus analogy with the elements inside the image. The experimental results were computed using 11 Landsat images over the city of Bucharest and surrounding areas. The LDA model enables us to discover a wide range of scene evolution classes based on the various dynamic behaviors of the land cover. The results are compared with the Corinne Land Cover map. However, this is not a validation method but one that adds static knowledge about the general usage of the analyzed area. In order to help the interpretation of the results, we use several studies on forms of relief, weather forecast, and very high resolution images that can explain the wide range of structures responsible for influencing the dynamic inside the resolution cell.

This paper describes research that seeks to supersede human inductive learning and reasoning in high-level scene understanding and content extraction. Searching for relevant knowledge with a semantic meaning consists mostly in visual human inspection of the data, regardless of the application. The method presented in this paper is an innovation in the field of information retrieval. It aims to discover latent semantic classes containing pairs of objects characterized by a certain spatial positioning. A hierarchical structure is recommended for the image content. This approach is based on a method initially developed for topics discovery in text, applied this time to invariant descriptors of image region or objects configurations. First, invariant spatial signatures are computed for pairs of objects, based on a measure of their interaction, as attributes for describing spatial arrangements inside the scene. Spatial visual words are then defined through a simple classification, extracting new patterns of similar object configurations. Further, the scene is modeled according to these new patterns (spatial visual words) using the latent Dirichlet allocation model into a finite mixture over an underlying set of topics. In the end, some statistics are done to achieve a better understanding of the spatial distributions inside the discovered semantic classes.

Urban areas around the world are rapidly changing in an unregulated manner and remote sensing is the most effective option for their monitoring and planning. Good modeling of urban areas means reliable translation of the scene semantics into an algorithmic language. The compression based image retrieval techniques are data driven. The intention of employing compression based image retrieval techniques is to exploit the compression properties of the objects and estimate the shared information between them. Fast compression distance (FCD) is the similarity metric used in a compression based image retrieval technique that can be applied on large datasets. FCD between any two objects can be computed using the sizes of their dictionaries (sequence of recurring patterns) extracted through compression with LZW algorithm and the intersection of their dictionaries. In this paper, it is proposed to assess high resolution Earth Observation data content based on data compression for understanding urban settlements.

Remote sensing platforms acquire huge amounts of data every day. As a result, large archives of data have been created. In order to provide access to this data efficient search and retrieval methods have to be developed, such as image information mining systems. In this paper we present a framework for an image information mining system using the Latent Dirichlet Allocation text-mining algorithm to provide a high-level semantic model of data, the search being performed in the LDA model space.

In this paper, a novel active learning approach and system incorporating multiple instance learning for SAR image mining and annotation is introduced. Based on a multiscale and hierarchial patch based image representation, a cascade classifier is learned at different levels. At each level of the hierarchy, a SVM classifier is trained based on active learning and the training sample propagation between different levels is achieved through Multiple Instance SVM (MI-SVM). Classification at the higher level is applied only to the positive patches obtained at the previous level, which can significantly reduce the burden of computation in the case of large data set. Performance has been evaluated through a large data set, which shows promising gain not only in accuracy but also in computation.

The main purpose of this study is to define for Synthetic Aperture Radar (SAR) data the primitive feature parameters, the incidence angle, and the orbit direction which can be used further for indexing and querying in the EO systems. The evaluation is done on the high resolution SAR data and the interpretation is realized automatically. In this paper, we propose to study and asses the behavior of the primitive feature extracted methods for images of the same scene with two look angles covering the min-max range of the sensor and with ascending / descending orbit looking. The tests are done on TerraSAR-X products Stripmap and high resolution Spotlight, specially and radiometrically enhanced covering the area of Berlin (Germany) and Ottawa (Canada). To identify the optimal primitive features, incident angle, and orbit direction the Support Vector Machine and as a measure of the classification accuracy the precision/recall were considered.

This paper presents an implementation of query by example in Earth Observation image archive using data compression-based approach. Data compression approach allows to exploit the compression properties of the objects and to estimate the shared information between them, this concept is extended to image retrieval for finding similar objects in the image archive. Our implementation is based on LZW algorithm for compressing the image content and extracting features of the images. The fast compression distance (FCD) is defined as a similarity metric in order to retrieve the most similar images. This tool is satisfactory implemented and tested using optical and SAR images.

A major drawback of classical SAR interferometry is its sensitivity to temporal and geometric effects, leading to a distortion of the final results. This problem can be solved by using multi-temporal techniques, like the Small Baseline Subset and Persistent Scatterers algorithms. Both use large datasets for monitoring land cover deformation, motion or for DEM generation. Despite the fact that using the Persistent Scatterers technique can lead to a maximization of the number of acquisitions used, the number of persistent points is greatly reduced by a decrease of the coherent values of the targets over time. In this article, the authors propose a method for the selection of temporal persistent points using image subsets that comply with the small baseline rule. This method is further used to increase the accuracy of deformation measurements over the subsets' time span.

In this paper will be applied concepts like amortized complexity or self-adjustment to binary search trees. Motivation comes from the fact that the search trees have multiples drawbacks. It will be developed and analyzed the splay tree, a self-adjusting form of binary search trees.

The recognition and analysis of moving targets in SAR images is a well-known topic in remote sensing. The public availability of high resolution SAR satellite images opened new perspectives for space-based small object recognition: traffic monitoring from space has matured into an established technique where the motion characteristics of ships and cars can be extracted routinely. A particular case of traffic monitoring is the search for moving trains in SAR images; here one can locate moving trains, determine their velocity, and identify the type and number of wagons. In the following, we concentrate on features to be extracted from TerraSAR-X images containing commuter trains in the area of Munich, Germany.

Very-high resolution (VHR) synthetic aperture radar (SAR) images from the last generation satellites such as TerraSAR-X and TanDEM-X exhibits special characteristics, especially in the urban-areas. Consequently, attention is needed on special considerations while developing algorithms for SAR image processing and its applications for automated interpretation. With automatic interpretation we refer to the information extraction and characterization for image categorization, retrieval, segmentation, automated target recognition etc.. In this article we focus our attention to the problem of SAR image categorization. The interest in SAR image categorization in VHR SAR (on the contrary to the pixel-based classification in low-resolution SAR images) has increased with enhanced resolution providing opportunity to carry out a more detailed analysis of targets and objects. SAR image categorization requires generation of a compact feature descriptor which accurately define the image content. A feature descriptor can be generated using `parametric' or `non-parametric' approaches in `image' or `image within a transformation space'. The objective of this article is to review some selected techniques for this purpose in form of a methodological classification. Qualitative assessment of selected algorithms is presented.

The new generation of spaceborne SAR instruments with meter or submeter resolution finds enormous applications for the observation of urban, industrial, in general of man-made scenes. Thus, targets are not any more observed in isolation, instead the groups of objects, e.g., house, bridge, and road, etc., need to be recognized in their spatial context. This paper proposes a feature extraction method for image patches in order to capture the spatial context. The method is based on the characteristics of the spectra of the SAR data, integrating radiometric, geometric, and texture properties of the SAR image patch. The method is demonstrated for TerraSAR-X High Resolution Spotlight data. To account for the spatial context in which a group of targets is located, it uses an image patch covering typically 200 × 200m2of the scene. A comparative evaluation of our descriptors and grey-level co-occurrence matrix (GLCM) texture features has been performed over a database of 6916 patches. The method allowed for the robust recognition of over 30 different scene classes, with precision between 50% and 93%. Numerical results show that our method is able to discriminate between scene classes better than GLCM texture parameters.

The advent of submeter-resolution synthetic aperture radar (SAR) images from satellites such as TerraSAR-X has given a new dimension to SAR image understanding. Even though emphasis is always on discovering automatic means of target characterization, visual exploration of targets and objects is the first step in many applications. While considering the complex-valued SAR images, visual inspection of the targets in an image may provide incomplete and misleading information, as sometimes two entirely different behaving objects look quite similar in SAR images. Thus, a need was felt to develop a methodology to support visual target recognition and analysis. In this letter, we present a method which looks into the complex-valued spectrum of SAR images, thus allowing a detailed physical interpretation of the scattering behavior of objects. The presented method is a joint time-frequency analysis method based on sublook decomposition. With the presented results, we emphasize the use of complex-valued SAR images for target characterization, the use of which is primarily restricted to polarimetric and interferometric applications as of now.

TELEIOS is a recent European project that addresses the need for scalable access to petabytes of Earth Observation data and the discovery and exploitation of knowledge that is hidden in them. TELEIOS builds on scientific database technologies (array databases, SciQL, data vaults) and Semantic Web technologies (stRDF and stSPARQL) implemented on top of a state of the art column store database system (MonetDB). We demonstrate a first prototype of the TELEIOS Virtual Earth Observatory (VEO) architecture, using a forest fire monitoring application as example.

Speckle hinders information in synthetic aperture radar (SAR) images and makes automatic information extraction very difficult. The Bayesian approach allows us to perform the despeckling of an image while preserving its texture and structures. This model-based approach relies on a prior model of the scene. This paper presents an evaluation of two despeckling and texture extraction model-based methods using the two levels of Bayesian inference. The first method uses a Gauss–Markov random field as prior, and the second is based on an auto-binomial model (ABM). Both methods calculate a maximum a posteriori and determine the best model using an evidence maximization algorithm. Our evaluation approach assesses the quality of the image by means of the despeckling and texture extraction qualities. The proposed objective measures are used to quantify the despeckling performances of these methods. The accuracy of modeling and characterization of texture were determined using both supervised and unsupervised classifications, and confusion matrices. Real and simulated SAR data were used during the validation procedure. The results show that both methods enhance the image during the despeckling process. The ABM is superior regarding texture extraction and despeckling for real SAR images.

In the context of image search and classification, we describe an active learning strategy that relies on the intrinsic data distribution modeled as a mixture of Gaussians to speed up the learning of the target class using an interactive relevance feedback process. The contributions of our work are twofold: First, we introduce a new form of a semi-supervised C-SVM algorithm that exploits the intrinsic data distribution by working directly on equiprobable envelopes of Gaussian mixture components. Second, we introduce an active learning strategy which allows to interactively adjust the equiprobable envelopes in a small number of feedback steps. The proposed method allows the exploitation of the information contained in the unlabeled data and does not suffer from the drawbacks inherent to semi-supervised methods, e.g. computation time and memory requirements. Tests performed on a database of high-resolution satellite images and on a database of color images show that our system compares favorably, in terms of learning speed and ability to manage large volumes of data, to the classic approach using SVM active learning.

In this paper, we describe an active learning scheme which performs coarse to fine testing using a multiscale patch-based representation of images to retrieve objects in large satellite image repositories. The proposed hierarchical top-down approach reduces step by step the size of the analysis window, eliminating each time large parts of the images considered as non-relevant. Unlike most object detection methods which requires large training sets and costly offline training, we use an active learning strategy to build a classifier at each level of the hierarchy and we propose an algorithm to propagate automatically the training examples from one level to the other.

The increasing number and resolution of earth observation (EO) imaging sensors has had a significant impact on both the acquired image data volume and the information content in images. There is consequently a strong need for highly efficient search tools for EO image databases and for search methods to automatically identify and recognize structures within EO images. In this paper, we present a concept for an earth observation image data mining system mixing an auto-annotation component with a category search engine which combines a generic image class search and an object detection feature. The proposed concept relies thus on three distinct components which are detailed successively: in the first part, we describe the auto-annotation component, in the second part, the generic category search engine and in the third part, the object detection tool. In the concluding part of the paper, we provide an insight into how these three components can be related to each other and used in a complementary way to arrive at a system which combines the advantages of both the auto-annotation systems and the category search engines.

The artifacts detection is a step of data cleaning process. The classical approach is to predict or determine the existence of defects, to model it, and then design a method to detect and correct them. This classical approach is for specific artifacts. The approach presented in this work is using complexity distortion theory to implement a more generic method, thus, this work will aim at developing parameter free methods able to automatically detect artifacts in EO images. We use the Kolmogorov Structure Function as approximation to the Rate-distortion curve and examine how the artifacts can have the same structure.

This paper presents a methodology for feature extraction from high resolution SAR image classification, using descriptors constructed from the complex SAR signal. The proposed data mining scheme aims at determining regions in the imaged scene which have similar content. Two complementary approaches are proposed, one making use of the single look complex data for feature extraction and the other based on the interferometric information available about the imaged scene. The features are derived from the estimated signal spectrum, in two stages. For the second stage, the model order is given by minimum number of components needed for classification and is estimated through the Akaike information criterion. Tests show that the proposed features allow for a robust recognition of 25 scene classes.

This work presents a new approximation for the Kolmogorov complexity of strings based on compression with smallest Context Free Grammars (CFG). If, for a given string, a dictionary containing its relevant patterns may be regarded as a model, a Context-Free Grammar may represent a generative model, with all of its rules (and as a consequence its own size) being meaningful. Thus, we define a new complexity approximation which takes into account the size of the string model, in a representation similar to the Minimum Description Length. These considerations result in the definition of a new compression-based similarity measure: its novelty lies in the fact that the impact of complexity overestimations, due to the limits that a real compressor has, can be accounted for and decreased.

This paper discusses methods and parameter settings that help to estimate texture in SAR images. In general, this is a difficult task for SAR images that are characterized by speckle noise and which span a wide range of pixel magnitudes. We applied Gauss Markov Random Field (GMRF) models and Enhanced Model Based Despeckling (EMBD) to 1 meter resolution amplitude images of the German TerraSAR-X mission. The results demonstrate that one can find appropriate parameter combinations that allow robust texture estimation even for different types of target areas.

This paper discusses the basic paradigm of how image information mining methods work in the field of remote sensing. To this end, we compare our approaches to the approaches being used in the world of multimedia; then we discuss the annotation specifics of remote sensing data and describe the different types of remote sensing data that we are faced with today. We conclude with a description of algorithmic alternatives and compare several competing systems that are currently in use. Finally, we provide an outlook of what we expect in the near future.

This paper presents two approaches for the analysis of multi-temporal and multisensory data analysis. The first approach focuses on a novel similarity measure, named Mixed Information, derived from information measures and assessment of its performance for multitemporal analysis in change detection. The second approach proposes the use of Kullback-Leibler divergence between marginal distributions, efficiently approximated by cumulant based expansion series. Comparison between mixed information and Kullback-Leibler divergence based change detection is performed to confirm the usefulness of joint metric for analyzing multitemproal and multisensory data. Experimental results obtained confirm justification of the mixed information in multitemporal analysis.

This paper presents a non-parametric modeling scheme for high resolution SAR data, based on Short Time Fourier Transform which is able to integrate the radiometrical and morphological properties of the data, for object recognition, scene and target indexing, addressing the problem of large data base queries and information retrieval.. The method is assessed by using a Bayesian Support Vector Machine image search engine based on a hierarchical learning model. The method allowed for the recognition of over 30 different classes, both homogeneous and heterogeneous urban objects with high levels of details. Qualitative and quantitative measures for evaluation are presented and discussed.

In this paper we discuss Gauss-Markov Random Field (GMRF) based on multiple sub-aperture decomposition method for the analysis of targets in complex-valued high-resolution SAR data. Gauss-Markov Random Field (GMRF) model with a quadratic energy function as a parametric analysis parameterizes the spectogram of the signal, whereas sub-aperture decomposition method exploits the holographic property of the spectrum at the cost of reducing resolution. This analysis helps to understand, characterize and analyze complex-valued SAR data and provides temptation to use complex-valued SAR data over detected data.

In this letter, our main objective is to demonstrate the capabilities of independent component analysis (ICA) for very high resolution satellite image characterization, particularly for geometrical structures contained in images from urban areas in comparison with natural landscapes. We propose a model based on ICA sources to index satellite images with a resolution of 0.6 m. An important part of this work is to study the effects of multiscaling and dimensionality behavior of ICA components. For evaluating the capabilities of our model, we design a classification method to distinguish images from a variety of natural and man-made scenes.

Earth observation image-understanding methodologies may be hindered by the assumed data models and the estimated parameters on which they are often heavily dependent. First, the definition of the parameters may negatively affect the quality of the analysis. The parameters could not be captured in all aspects, and those resulting superfluous or not accurately tuned may introduce nuisance in the data. Furthermore, the diversity of the data, as regards sensor type, spatial, spectral, and radiometric resolution, and the variety and regularity of the observed scenes make it difficult to establish enough valid and robust statistical models to describe them. This letter proposes algorithmic information theory-based analysis as a valid solution to overcome these limitations. We will present different applications on satellite images, i.e., clustering, classification, artifact detection, and image time series mining, showing the generalization power of these parameter-free data-driven methods based on the computational complexity analysis.

With the launch of the German TerraSAR-X system in June 2007, a new generation of high-resolution spaceborne synthetic aperture radar (SAR) data is available, which should facilitate the interpretation of urban environments. Our overall objective in this letter is to provide a semiautomatic tool for urban area interpretation using SAR data. We propose in this letter to fuse different automatic object extractors in order to provide more reliable pieces of interpretation. Our fusion is a coarse-to-fine approach. First, a segmentation of the image is performed to partition the scene into regions having similar properties. The second step consists in detecting bright and dark linear structures which are, in general, linked to the presence of buildings and roads (main classes in urban areas), respectively. The last step gives the final image interpretation using contextual knowledge. Evaluation of the proposed approach in mapping urban areas was carried out using real TerraSAR-X data over the city of Las Vegas in the U.S.

Synthetic aperture radar (SAR) images are affected by multiplicative noise called speckle. This noise makes automatic image classification and image interpretation difficult. Thus, many methods have been developed to remove speckle from SAR images while preserving the useful information of the scene such as texture and geometry. In this letter, a comparison between three different despeckling methods based on a Bayesian approach and Gibbs random fields is made. The used methods are Gauss–Markov random field (GMRF) and autobinomial modeling, which operate in the image domain, and the GMRF approach, which operates in the wavelet domain. Our methods are evaluated with synthetic and real SAR data (TerraSAR-X images). The experimental results show that, with these three methods, the speckle is well removed while structures are preserved; quantitative measures show that the autobinomial method provides the best smoothness and sharpness criteria in real SAR data, while the wavelet-based method generates the smallest bias.

This letter presents the despeckling of synthetic aperture radar (SAR) images within the bandelet and contourlet domains. A model-based approach is presented for the despeckling of SAR images. The speckle-reduced estimate is found using the first-order Bayesian inference, and the best model's parameters are estimated using the second-order Bayesian inference. Synthetic and real images are used for evaluating the qualities of the despeckling methods. The experimental results showed that the combination of Bayesian inference and bandelet transform outperforms the contourlet-based despeckling algorithm using synthetic data and objective measurements.

In this letter, we are interested in the annotation of large satellite images, using semantic concepts defined by the user. This annotation task combines a step of supervised classification of patches of the large image and the integration of the spatial information between these patches. Given a training set of images for each concept, learning is based on the latent Dirichlet allocation (LDA) model. This hierarchical model represents each item of a collection as a random mixture of latent topics, where each topic is characterized by a distribution over words. The LDA-based image representation is obtained using simple features extracted from image words. We then exploit the capability of the LDA model to assign probabilities to unseen images, in order to classify the patches of the large image into the semantic concepts, using the maximum-likelihood method. We conduct experiments on panchromatic QuickBird images with 60-cm resolution. Taking into account the spatial information between the patches shows to improve the annotation performance.

Generally, the repository of a remote sensing system needs a capacity of hundreds of terabytes to archive original and processed data. There are several satellites whose downlink rates require ingesting over 100 GB into the archive each day. In order to process such a huge volume of data, to classify them, to index generated classes in the database, and to provide a quick access to stored information at real time, an image information mining system is developed. In this letter, we present the design of a system architecture for processing data from very large image archives.

This letter presents the despeckling of single-look complex (SLC) synthetic aperture radar (SAR) images using nonquadratic regularization. The objective function consists of an image model, a gradient, and a prior model. The Huber–Markov random field (HMRF) models the prior. A numerical solution is achieved through extensions of half-quadratic regularization methods using complex-valued SAR data. The proposed method using the HMRF prior together with nonquadratic regularization shows the superior results on SLC synthetic and actual SAR images.

Information content and compression are tightly related concepts that can be addressed by classical and algorithmic information theory. Several entities in the latter have been defined relying upon notions of the former, such as entropy and mutual information, since the basic concepts of these two approaches present many common tracts. In this work we further expand this parallelism by defining the algorithmic versions of cross-entropy and relative entropy (or Kullback-Leiblerdivergence), two well-known concepts in classical information theory. We define the cross-complexity of an object x with respect to another object y as the amount of computational resources needed to specify x in terms of y, and the complexity of x related to y as the compression power which is lost when using such a description for x, with respect to its shortest representation. Since the main drawback of these concepts is their uncomputability, a suitable approximation based on data compression is derived for both and applied to real data. This allows us to improve the results obtained by similar previous methods which were intuitively defined.

In this paper, we present a semi-supervised method for auto-annotating image collections and discovering unknown structures among them. The approach relies on the existence of only a small training database of annotated examples. First, a fully-supervised algorithm using annotated samples is presented. Next, we introduce a semi-supervised procedure which allows us to incorporate unannotated samples and to infer the existence of unknown structures, that is, the existence of new image classes which are not represented in the training database. Finally, we present experimental results from a database of satellite images and briefly mention the possibility of reusing the presented approach as a basis for more complex systems such as Content Based Image Retrieval (CBIR) systems.

Many fawns and other wild animals are killed by mowing machines every year. To prevent them from being killed or injured, a sensor system is being developed to detect the fawns hidden in meadows under mowing. Beside a microwave radar system, two cameras (thermal infrared and RGB) take a picture at the mower's current location. This contribution focuses on the compression-based algorithm that will be adopted to detect the locations containing a fawn hiding in the grass: such approach, being parameter-free, allows performing a fully unsupervised clustering by exploiting the intrinsic properties of data compression to estimate the amount of shared information between two images.

In this paper a new algorithm for interferometric phase restoration is presented. Firstly, a continuous framework for anisotropic phase diffusion is stated. A tensorial based metric allows directional control. The periodic continuous structure of the phase representation is accounted for. Secondly, this framework is adapted for interferometric phase filtering. Progressive re-estimation of directionality avoids directional bias. Isotropy and anisotropy are adaptively combined with a constant overall diffusion rhythm, so that the degree of regularization is the same regardless of the underlying topography. Robust estimation minimizes the spread of outliers. Results on both synthetic and TerraSAR-X data are provided.

In this paper a new Bayesian algorithm for interferometric phase restoration is presented. Based on a non-linear anisotropic extension of Orientation Diffusions, the inherent directionality of the fringe structure is introduced into its prior model. It also accounts for the periodicity of the phase representation. A fidelity term derived from the anisotropic metrics and the InSAR phase statistics deviates diffusion towards the acquired phase value. It acts as an adapted likelihood of the diffused phase. Hence, phase restoration is a trade-off between directionality and reconstruction fidelity, prior and likelihood. Results are provided on a High Resolution Spotlight scene acquired by TerraSAR-X.

High resolution remote sensing SAR images — such as the image data acquired by the German TerraSAR-X mission — contain a variety of details that have to be extracted by automated processing in order to fully exploit and understand the image content. In particular, the interpretation of man-made structures that are typical of built-up or agricultural areas poses a number of challenges including parameterized image focusing during routine processing, careful despeckling, descriptor and feature extraction, and final classification including specific scattering and 3D effects. Therefore, we propose a set of general sequential as well as dedicated application-dependent processing steps that allow user-oriented classification of high resolution SAR images. We will also report on actual classification results and experiences.

In this paper we compare parametric and non-parametric method for the analysis of complex valued high-resolution SAR data. Gauss-Markov Random Field (GMRF) model with a quadratic energy function as a parametric analysis parameterizes the spectogram of the signal, whereas nonlinear short time Fourier transform (STFT) method, the method based on time frequency analysis (TFA) as a non-parametric approach exploits the signal's non-stationarity in the time-frequency domain for information extraction. This comparative analysis helps to understand, characterize and analyze complex valued SAR data.

The ill-defined nature of the segmentation problem makes the selection of the optimal image partition difficult. One can characterize image segmentation as an attempt to find the best possible representation of a data set using a certain number of ldquoobjects.rdquo This can be regarded as data information compression, resulting in the distortion of the original values. Data sets are well represented when the correct number of regions is chosen. The concept behind this approach is similar to the main problem of rate distortion theory: A finite set of code words is chosen to approximate the numbers or source symbols as well as possible. In our approach, the number of regions is equivalent to the number of code words. The mean of a region provides canonical representation of respective group members, and the distortion function is the mean-square error assuring a good evaluation method for image segmentation.

This paper proposes a new-wavelet-based synthetic aperture radar (SAR) image despeckling algorithm using the sequential Monte Carlo method. A model-based Bayesian approach is proposed. This paper presents two methods for SAR image despeckling. The first method, called WGGPF, models a prior with Generalized Gaussian (GG) probability density function (pdf) and the second method, called WGMPF, models prior with a Generalized Gaussian Markov random field (GGMRF). The likelihood pdf is modeled using a Gaussian pdf. The GGMRF model is used because it enables texture parameter estimation. The prior is modeled using GG pdf, when texture parameters are not needed. A particle filter is used for drawing particles from the prior for different shape parameters of GG pdf. When the GGMRF prior is used, the particles are drawn from prior in order to estimate noise-free wavelet coefficients and for those coefficients the texture parameter is changed in order to obtain the best textural parameters. The texture parameters are changed for a predefined set of shape parameters of GGMRF. The particles with the highest weights represents the final noise-free estimate with corresponding textural parameters. The despeckling algorithms are compared with the state-of-the-art methods using synthetic and real SAR data. The experimental results show that the proposed despeckling algorithms efficiently remove noise and proposed methods are comparable with the state-of-the-art methods regarding objective measurements. The proposed WGMPF preserves textures of the real, high-resolution SAR images well.

Many methodologies and similarity measures based on data compression have been recently introduced to compute similarities between general kinds of data. Two important similarity indices are the normalized information distance (NID), with its approximation normalized compression distance (NCD), and the pattern recognition based on data compression (PRDC). At first sight NCD and PRDC are quite different: the former is a direct metric while the latter is a methodology which computes a compression distance with an intermediate step of encoding files into texts. In spite of this, it is possible to demonstrate that they are both based on estimates of Kolmogorov complexities (when this is known for the former but not for the latter). Finally, this results in the definition of a new measure: the model conditioned data compression based similarity measure (McDCSM), which is a modified version of PRDC, and is the topic of this paper.

Images may contain blemishes or artificial structures which come from the processing or directly from the sensors, that decrease the quality of the images and can lead to analysis and interpretation problems. We are interested in detecting these defects automatically. The aim of our method will be to look for artifacts as too regular or too irregular image structures that will be detected using the basic properties of Kolmogorov complexity.

This paper proposes complexity based analysis as a valid alternative to classic image analysis methodologies for Earth Observation imagery, which are heavily dependant on the assumed data models. These methods are totally model-free and data-driven, and may be successfully employed for image classification and indexing, regardless of spatial and radiometric resolution of the scene and sensor type.

With the launch of the German TerraSAR-X system in June 2007, a new generation of high-resolution spaceborne synthetic aperture radar (SAR) data is available; which should facilitate the interpretation of urban environments. This article proposes a new automatic tool for geometric and topological urban areas characterization. Our approach is divided into three main steps. First, a bright linear structures detector is applied to extract the geometrical information. Then, a graph-based spatial characterization is used to model the topological relationships between the different detected bright pixels (nodes). Next, a classification by examining the profile of the distributions of the angles between neighboring nodes, is performed in order to label the linked structures. Evaluations of the proposed approach in characterizing the geometry and topology of urban areas were carried out using TerraSAR-X data over three different cities: Las Vegas in the United States, Paris in France and Cairo in Egypt.

While typical remote sensing imaging instruments produce more and more data, what we miss today are reliable tools for automated information extraction form these images. In the following, we propose a so-called Content Map, a novel Earth Observation value adding product. Basically, it comprises several class files and a viewer showing the different classes of land use and objects contained in the corresponding image data. In order to avoid processing delays, the class files have to be generated in an unsupervised mode as a real time product; thus, interactive user interactions have to be limited to training and testing intervals. As typical examples we use image data of the German TerraSAR-X mission that produces SAR image data in a variety of different modes.

Remote sensing Earth Observation images may contain blemishes or artificial structures generated by the processing or directly by the sensors. These artifacts decrease the quality of the images and may lead to analysis and interpretation problems. We are interested in detecting these defects automatically. A visual analysis of such artifacts brings us to the conclusion that they constitute a local modification of the complexity of the image. Therefore, this paper will give a complexity-based approach for the detection of these defects. Our approach will comprise two different methods: in a first approach, the artifacts will be compared with given examples based on a complexity comparison. Due to some weaknesses of this method a second approach will be defined which will not depend on given examples but will take into consideration the local complexity context of the images.

Compared to conventional optical images, the classification of remote sensing SAR images represents a rather difficult task. As a rule, the various SAR imaging and product options, the high dynamic range of SAR images, and the presence of speckle noise prevent us from obtaining robust classification results. In the following, we try to circumvent these difficulties by proper pre-processing and despeckling of high resolution SAR images. Our approach aims at adapting the radiometry and the resolution of the scenes to the optimal target recognition capabilities of a classification algorithm. To this end, we have to apply systematic adaptations that provide optimized multilooking of our input data. Then these adapted data can be fed into a scene classification system.

In this paper we present two algorithms for information extraction from Single Look Complex (SLC) Synthetic Aperture Radar (SAR) images. The first algorithm is based on Tikhonov regularization with Total Variation (TV) and a Point-Based Feature (PBF) term. Based on the equivalence of Tikhonov and the Bayesian estimate, the second algorithm is a Maximum A Posteriori (MAP) estimation with a complex-valued Gauss-Markov Random Field (GMRF) in addition to the TV prior. The first algorithm produces a despeckled image preserving fine details and texture. The second algorithm gives a denoised image and in addition the estimated feature parameter vector θ.

Earth observation optical images may contain blemishes or artificial structures that come from processing or directly from the sensors. These artifacts decrease the quality of the images and can lead to analysis and interpretation problems. We are interested in detecting these defects automatically. We develop an approach based on an information-theoretic analysis of the image formation process, which is inspired by methods already used in domains dealing with similar problematics like image quality or information hiding. In this letter, we propose a method based on a rate distortion analysis, exploiting the local regularity properties of the image.

This paper addresses the problem of building an index of compressed object databases. We introduce an informational similarity measure based on the coding length of two part codes. Then, we present a methodology for compressing the database by taking into account interobject redundancies and by using the informational similarity measure. The method produces an index included in the code of the data volume. This index is built such that it contains the minimal sufficient information to discriminate the data-volume objects. Then, we present an optimal two-part coder for compressing spatio-temporal events contained in satellite image time series (SITS). The two-part coder allows us to measure similarity and then to derive an optimal index of SITS spatio-temporal events. The resulting index is representative of the SITS information content and enables queries based on information content.

This paper addresses the problem of building a compact representation of objects which enables to define a similarity measure. The key idea is to provide a mean to extract and represent the relevant information of an object. By integrating the minimum length description to an informational similarity metric, a new informational measure is proposed based on models. First, we make the strong assumption that for any object x there exists a model Mxsuch that K(x) = K(x | Mx) + K(Mx)

With the increase of synthetic aperture radar (SAR) sensor resolution, SAR images could include a large variety of interesting real man-made structures. Therefore, a more detailed analysis and a finer description of SAR images of urban areas are needed for a better understanding of the scene. Nevertheless, recognizing scenes using high resolution SAR images requires the capability to identify relevant signal signatures (called also descriptors), depending on variable image acquisition geometry, arbitrary objects poses and configurations. Among feature extraction methods, we propose to use principal components analysis (PCA) and/or independent components analysis (ICA), in order to exploit deeper the nature of SAR signatures. In this paper, both a description of our work and a presentation of our preliminary classification performance results will be provided.

The last two decades showed an important development of satellite imagery with past and present satellites acquiring enormous volumes of data. Meanwhile, the quality of the acquired images increased permitting the recording of high resolution images (0.6÷2.5 meters/pixel) in multispectral bands. Thus, both the data volume and the information detail increase dramatically. Consequently, new methods and tools to access and interpret Earth Observation (EO) images are needed. The present paper presents a semantic search engine for High Resolution (HR) EO images based on a hierarchical information model of satellite image contents. To face the potentially ambiguos meaning of image structures depending on their contextual understanding, the search engine uses Bayesian inference to learn categories and a Support Vector Machine (SVM) classifier to assign semantics. The categories are grouping and memorising the semantics of image structures, facilitating their recognition in various contexts. Also the generation of categories helps learning from a small training data set (i.e. image examples); thus, the method is useful for the exploitation of very large data volumes. The concept has enhanced inferred power, therefore optimising the Human Machine Communication (HMC), which is enhanced with learning / unlearning functions.

Nowadays huge volumes of Earth Observation data are available and increasing every day. Users are faced with the problem of having to extract and interpret information in large volumes of data, with the additional problem of images being annotated only by simple descriptors. Therefore, in this paper we present the Knowledge-centered Earth Observation system which offers interactive probabilistic information mining, among other functionalities, and two Support Vector Machine based Feature Extraction services integrated in the system.

Modern imaging sensors, especially those aboard satellites, continuously deliver enormous amounts of data. The widespread of meter resolution images, is not only exploding the volumes of acquired data but also brings a new dimension in the image detail, thus growing the information content. These represent typical cases, where users need automated tools to discover, explore and explain the contents of large image databases. There is a strong need to build up applications that help the user in image interpretation task, applications that permit to query the archives in content based mode, without having to know all the information contained in the images at signal level. We propose in this article, a synergy between stochastic modelling, knowledge discovery, and semantic representation. To do that, we associate semantic labels to a combination of primitive image features. The user-defined semantic image content interpretation is linked with Bayesian networks to a completely unsupervised classification. This new paradigm for the interaction with EO archives can provide several applications for users coming from different domains, as change detection, agricultural field classification, environment monitoring, atmosphere effects or urbanization.

The paper presents two algorithms for texture primitive feature extraction on Single Look Complex (SLC) and Polarimetric Synthetic Aperture Radar (PolSAR) SLC data. We assume the data to be modeled by a Gauss-Markov Random Field (GMRF): a complex GMRF model for characterizing the spatial correlation in SLC data and an extension of the model for inter-band correlation characterization. The complex GMRF characterizes the spatial relationship of a two-dimensional complex signal, i.e. SLC SAR data. The extended model characterizes the spatial interaction and the inter-band pixels correlation between the polarimetric complex channels. The Bayesian approach permits to deal with model fitting and selection in a direct way. The results are presented on a polarimetric E-SAR L band scene of Mannheim, Germany.

The eight papers in this special section describe some of the recent advances made in the field of image information mining (IIM) for Earth Observation in the development of tools, methods, and applications.

In this paper, a wavelet-based speckle-removing algorithm is represented and tested on synthetic aperture radar (SAR) images. The SAR image is first transformed using a dyadic wavelet transform. The noise in the wavelet-transformed image is modeled as an additive signal-dependent noise with Gaussian distribution. The distribution of a noise-free image in a wavelet domain is modeled as a generalized Gauss–Markov random field (GGMRF). An unsupervised stochastic model-based approach to image denoising is represented. If the observed area is homogeneous, the parameters of the Gaussian distribution and GGMRFs are estimated from incomplete data using mixtures of wavelet coefficients. An expectation–maximization algorithm is used to estimate the parameters of both noisy and noise-free images. The unknown parameters are estimated using image and noise models that are defined in the wavelet domain for heterogeneous areas. Different inter- and intrascale dependences of wavelet coefficients were used to estimate the unknown parameters. The represented wavelet-based method efficiently removes noise from SAR images.

Satellite image time series (SITS) consist of a time sequence of high-resolution spatial data. SITS may contain valuable information, but it may be deeply hidden. This paper addresses the problem of extracting relevant information from SITS based on the information-bottleneck principle. The method depends on suitable model selection, coupled with a rate-distortion analysis for determining the optimal number of clusters. We present how to use this method with the Gauss-Markov random fields and the autobinomial random fields model families in order to characterize the spatio-temporal structures contained in SITS. Experimental results on synthetic data and SITS from SPOT demonstrate the performance of the proposed methodology

Presently Earth observation (EO) satellites acquire huge volumes of high resolution images very much over-passing the capacity of the users to access the information content of the acquired data. Thus, in addition to the existing methods for EO data and information extraction, new methods and tools are needed to explore and help to discover the information hidden in large EO image repositories. This article presents a categorisation based Relevance Feedback (RF) search engine for EO images repositories The developed method is presented as well results obtained for a SPOT5 satellite image database.

This letter presents synthetic aperture radar (SAR) image despeckling using dyadic wavelet transform. Maximum a posteriori (MAP) estimation is used to despeckle a SAR image in the wavelet domain. A wavelet transformed speckle-free image is approximated with a Gauss–Markov random field, and a Gaussian model is chosen to approximate speckle in the wavelet domain. A speckle-free wavelet coefficient is estimated with Bayesian inference using image and noise model parameters, which produce the highest evidence. The experimental results showed that the despeckling algorithm removes speckle noise in the homogeneous areas better than the state-of-the-art methods, which operate in the wavelet and image domain. The proposed method is very simple and computationally not demanding.

We present a complete processing line to generate an object based description of optical remote sensing (RS) images. A segmentation algorithm is used to generate a partition of regions and simplify the volume of data. Results are still at the pixel level. Based on topology analyses, a dynamical algorithm is proposed to retrieve, extract the segmented regions and encode them in a tree structure which describes their topological relations (adjacencies, inclusions). The overall collected information constitutes a consistent and independent database, generated efficiently on standard workstation. Many applications are possible, such as content based image retrieval, image description and compression, object classification or image-object fusion. A scenario is presented to emphasize interest of the method in the case of 3D visualization enhancement: image-objects are integrated on elevation data (digital elevation models, DEM) in order to generate more realistic rendering.

In this paper a wavelet based method for SAR data denoising and compression is presented. An unsupervised stochastic model based approach to image denoising is presented. SAR image is modeled in wavelet domain Gauss Markov random field and noise is considered as Gaussian with unknown variance. The parameters are estimated from incomplete data using mixtures of wavelet coefficients, and expectation maximization algorithm. The expectation maximization algorithm is used to efficiently compute a maximum a posteriori estimate. Observed wavelet coefficient is estimated using inter and intra scale of wavelet coefficients to estimate image and noise model parameters. Presented wavelet based method efficiently removes noise from SAR images. The second step is to design an entropy coder that efficiently codes despeckled image. The texture parameters obtained at the despeckling stage are used in the compression process. The image coder is tested on X-SAR data with and achieves comparable compression results with the wavelet based state-of-the art coders for SAR data compression.

We present an algorithmic protocol for the evaluation of a content-based remote sensing image information mining system. In order to provide users fast access to the content of large image databases, the system is composed of two main modules. The first includes computationally intensive algorithms for off-line data ingestion in the archive, image feature extraction, and indexing. The second module consists of a graphical man-machine interface that manages the information fusion for interactive interpretation and the image information mining functions. According to the system architecture, the proposed evaluation methodology aims to determine the objective technical quality of the system and includes subjective human factors as well. Since the query performance of a content-based image retrieval system mainly depends on the datasets stored in the archive, we first analyze the complexity of image data. Then, we determine the accuracy of the interactive training that can be considered as a supervised Bayesian classification of the entire archive. Based on the stochastic nature of user-defined cover types, the system retrieves images using probabilistic measurements. The information quality of the queried results is measured by target and misclassified images, precision and recall, and the probability to forget and to overretrieve images. Since the queried images are the result of a number of interactions between user and system, we analyze the man-machine communication dialogue and the system operation, too. Finally, we compare the objective component of the evaluation protocol with the users' degree of satisfaction to point out the significance of the computed measurements.

The progress in information retrieval, computer vision, and image analysis makes it possible to establish very complete bases of algorithms and operators. A specialist in remote sensing or image processing now has the tools that allow him, at least in theory, to configure applications solving complex problems of image understanding. However, in reality, earth observation (EO) data analysis is still performed in a very laborious way at the end of repeated cycles of trial and error. To overcome this, we proposed a novel advanced remote sensing information processing system knowledge-driven information mining (KIM). KIM is based on human-centered concepts (HCCs), which implements new features and functions allowing improved feature extraction, search on a semantic level, the availability of collected knowledge, interactive knowledge discovery, and new visual user interfaces. We assess the HCC methodology for solving several difficult tasks in EO image interpretation, using a broad variety of sensor data, from meter-resolution synthetic aperture radar and optical images to hyperspectral data.

During the last decades, satellites have acquired incessantly high-resolution images of many Earth observation sites. New products have arisen from this intensive acquisition process: high-resolution satellite image time-series (SITS). They represent a large data volume with a rich information content and may open a broad range of new applications. This paper presents an information mining concept which enables a user to learn and retrieve spatio-temporal structures in SITS. The concept is based on a hierarchical Bayesian modeling of SITS information content which enables us to link the interest of a user to specific spatio-temporal structures. The hierarchy is composed of two inference steps: an unsupervised modeling of dynamic clusters resulting in a graph of trajectories, and an interactive learning procedure based on graphs which leads to the semantic labeling of spatio-temporal structures. Experiments performed on a SPOT image time-series demonstrate the concept capabilities.

This paper presents a system to integrate digital elevation model (DEM) enhancement and Earth Observation (EO) image analysis for realistic three-dimensional (3-D) rendering applications. There is an increasing interest in interferometric synthetic aperture radar (InSAR) data, principally due to the availability of the nearly global Shuttle Radar Topography Mission coverage. To remove artifacts and noise from an InSAR DEM, a nonstationary Bayesian filtering is applied that preserves structural information. Land-cover or man-made structures are easily recognized in an optical image. The corresponding geometry encapsulated in the DEM differs from our implicit perception and generally leads to unrealistic 3-D rendering. To improve this, DEM regularization is achieved using only the visualization dataset (optical image and DEM). It consists of extracting relevant information from the optical image and integrate them in the filtered DEM. To gather image information, an object-based description of large optical EO images is obtained in two stages 1) an image is segmented to create a partition of regions and 2) a novel dynamical algorithm is proposed to extract the regions and encode them in a tree structure. Regions are modeled by objects primitives stored in a database. Spatial relationships between regions are reflected by the presented tree of regions. Using the object-based description generation, structures to be integrated into the DEM are interactively selected and classified among a set of user-thematic. Each thematic is associated with a corresponding elevation modeling and enables to estimate the region's 3-D structure. The proposed object line processing provides more realistic 3-D visualizations.

Very large volumes of heterogenous data, like multimedia, Earth observation images, scientific and engineering measurements, for instance, are continuously generated and stored. A typical case is the field of Earth observation. The widespread availability of high resolution images does not only explore the volumes of data, but also brings order at magnitude in the image detail, thus enormously increasing the information content. However, today's concepts and technologies are still limited in communicating the information content to people for use in real life applications. In this paper, we overview a new concept for knowledge-driven image information mining (KIM) and both analyze and evaluate it from the perspective of human-machine communication. The KIM concept enables the information communication from a very large image repository to users via the Internet. The communication is at a semantic level of representation and is adapted to the user's conjecture.

The interpretation of EO data requires not only data and/or information fusion for better understanding of Earth cover structures, but, at a higher level, needs the aggregation with existing bodies of knowledge specific to the application fields.

A visual information mining concept is proposed for spatio-temporal patterns discovery in remotely sensed image time-series (ITS). An information theory framework is adopted to first model information content. It results in the inference of a relevant directed graph characterizing ITS. Then the user conjecture is modeled via visual information representations: similarity measures between sub-graphs, which represents spatio-temporal events are derived and included in an interactive learning and probabilistic retrieval procedure of user-specific event-types. The present concept for ITS mining is demonstrated on multitemporal SPOT data.

To investigate the limits and merits of information extraction from a single high-resolution synthetic aperture radar (SAR) backscatter image, we introduce a model-based algorithm for the automatic reconstruction of building areas from single-observation meter-resolution SAR intensity data. The reconstruction is based on the maximum a posteriori estimation by Monte Carlo methods of an optimal scene that is modeled as a set of mutually interacting Poisson-distributed marked points describing parametric building objects. Each of the objects can be hierarchically decomposed into a collection of radiometrically and geometrically specified object facets that in turn get mapped into data features by ground-to-range projection and inverse Gaussian statistics. The detection of the facets is based on a likelihood ratio. Results are presented for airborne data with resolutions in the range of 0.5-2 m on urban scenes covering agglomerations of buildings. To achieve robust results for building reconstruction, the integration with data from other sources is needed.

In this paper, a dynamic scene understanding concept is proposed and applied on multispectral image time series. Information mining enables the explorations and discovery of spatio temporal patterns localized in given spatio temporal windows. With this in mind, a hierarchical information representation is developed. It comprises different levels in which the data is modeled so that the relevant information is transmitted through the architecture, according to a query and to several assumptions made on the models employed. There are mainly four components: feature extraction, reduction of dimensionality, clustering and interactive exploration.

Several signal processing techniques are presented and evaluated to filter/enhance SAR digital elevation models (DEMs). The results are compared to a topographic digital terrain model (DTM) in the context of 3D visualization and real-time rendering. Through the DLR X-SRTM DEM, the interest of InSAR data for such applications is illustrated.

The exploitation of metric resolution SAR data for the reconstruction of the structure of the observed scenes poses specific problems related both to the complexity of acquired scene details and to the peculiarities of the SAR acquisition system. On the one hand, much more complexity is transferred through the system from the scene into the data: new kinds of complex man-made scene objects are acquired. Layover and shadowing and responses from single scatterers tend to dominate the data. On the other hand, multiple signal reflections, sidelobe effects, radiometric pollution and many other effects related to the increased resolution of the system have to be taken into account. We show how, by properly modelling in stochastic terms the peculiarities of both the acquisition system and of the scene and by composing them in a Bayesian framework, new methods are developed that allow the reconstruction of the imaged structures from SAR data. Particular interest is devoted to the application of the developed algorithms in urban environments on data resolutions ranging from a few metres to fifty centimetres.

In this paper, we demonstrate the concepts of a prototype of a knowledge-driven content-based information mining system produced to manage and explore large volumes of remote sensing image data. The system consists of a computationally intensive offline part and an online interface. The offline part aims at the extraction of primitive image features, their compression, and data reduction, the generation of a completely unsupervised image content-index, and the ingestion of the catalogue entry in the database management system. Then, the user's interests-semantic interpretations of the image content-are linked with Bayesian networks to the content-index. Since this calculation is only based on a few training samples, the link can be computed online, and the complete image archive can be searched for images that contain the defined cover type. Practical applications exemplified with different remote sensing datasets show the potential of the system.

We present a framework for scene understanding from interferometric synthetic aperture radar data that is based on Bayesian machine learning and information extraction and fusion. A generic description of the data in terms of multiple models is automatically generated from the original signals. The obtained feature space is then mapped to user semantics representing urban scene elements in a supervised step. The procedure is applicable at multiple scales. We give examples of urban area classification and building recognition of Shuttle Radar Topography Mission data and of building reconstruction from submetric resolution Intermap data.

The hierarchical segmentation (HSEG) algorithm is an approach for producing high quality, hierarchically related image segmentations. The VisiMine image information mining system utilizes clustering and segmentation algorithms for reducing visual information in multispectral images to a manageable size. The project discussed herein seeks to enhance the VisiMine system through incorporating hierarchical segmentations from HSEG into the VisiMine system.

In order to apply a statistical approach to the classification of multisource remote-sensing data, one of the main problems to face lies in the estimation of probability distribution functions. This problem arises out of the difficulty of defining a common statistical model for such heterogeneous data. A possible solution is to adopt nonparametric approaches, which rely on the availability of training samples without any assumption about the related statistical distributions. The purpose of this paper is to investigate the suitability of the concept of dependence trees for the integration of multisource information through estimation of probability distributions. First, this concept, introduced by Chow and Liu (1968), is used to provide an approximation of a probability distribution defined in an N-dimensional space by a product of N-1 probability distributions defined in two-dimensional (2-D) spaces; this approximation corresponds, in terms of graph theoretical interpretation, to a tree of dependence. For each land cover class, a dependence tree is generated by minimizing an appropriate closeness measure. Then, a nonparametric estimation of the second-order probability distributions is carried out through the Parzen window approach, based on the implementation of 2-D Gaussian kernels. In this way, it is possible to reduce the complexity of the estimation, while capturing a significant part of the interdependence among variables. A comparison with other multisource data fusion methods, namely, the multilayer perceptron (MLP) method, the k-nearest neighbor (k-NN) method, and a Bayesian hierarchical classifier (BHC), is made. Experimental results obtained on multisensor [airborne thematic mapper (ATM) and synthetic aperture radar (SAR)] and multisource (experimental synthetic aperture radar (E-SAR) and a textural feature) data sets show that the proposed fusion method based on dependence trees is able to provide a classification accuracy similar to those of the other methods considered, but with the advantage of a reduced computational load.

This paper deals with the automatic extraction of the road network in dense urban areas using a few-meters-resolution synthetic aperture radar (SAR) images. The first part presents the proposed method, which is an adaptation of previous work to the specific case of urban areas. The major modifications are 1) the clique potentials of the Markov random field that extracts the road network are adapted and 2) a multiscale framework is used. Results on shuttle mission and aerial SAR images with different resolutions are presented. The second part is dedicated to road extraction combining two SAR images taken with different flight directions (orthogonal and antiparallel passes), and the obtained improvement is analyzed.

The authors present a concept of interactive learning and probabilistic retrieval of user-specific cover types in a content-based remote sensing image archive. A cover type is incrementally defined via user-provided positive and negative examples. From these examples, the authors infer probabilities of the Bayesian network that link the user interests to a pre-extracted content index. Due to the stochastic nature of the cover type definitions, the database system not only retrieves images according to the estimated coverage but also according to the accuracy of that estimation given the current state of learning. For the latter, they introduce the concept of separability. They expand on the steps of Bayesian inference to compute the application-free content index using a family of data models, and on the description of the stochastic link using hyperparameters. In particular, they focus on the interactive nature of their approach, which provides instantaneous feedback to the user in the form of an immediate update of the posterior map, and a very fast, approximate search in the archive. A java-based demonstrator using the presented concept of content-based access to a test archive of Landsat TM, X-SAR, and aerial images are available over the Internet [http:/www.vision.ee.ethz.ch/-rsia/ClickBayes].

Basic textures as they appear, especially in high resolution SAR images, are affected by multiplicative speckle noise and should be preserved by despeckling algorithms. Sharp edges between different regions and strong scatterers also must be preserved. To despeckle images, the authors use a maximum aposteriori (MAP) estimation of the cross section, choosing between different prior models. The proposed approach uses a Gauss Markov random field (GMRF) model for textured areas and allows an adaptive neighborhood system for edge preservation between uniform areas. In order to obtain the best possible texture reconstruction, an expectation maximization algorithm is used to estimate the texture parameters that provide the highest evidence. Borders between homogeneous areas are detected with a stochastic region-growing algorithm, locally determining the neighborhood system of the Gauss Markov prior. Smoothed strong scatterers are found in the ratio image of the data and the filtering result and are replaced in the image. In this way, texture, edges between homogeneous regions, and strong scatterers are well reconstructed and preserved. Additionally, the estimated model parameters can be used for further image interpretation methods.

Automatic interpretation of remote-sensing (RS) images and the growing interest for query by image content from large remote-sensing image archives rely on the ability and robustness of information extraction from observed data. In Parts I and II of this article, the authors turn the attention to the modern Bayesian way of thinking and introduce a pragmatic approach to extract structural information from RS images by selecting from a library of a priori models those which best explain the structures within an image. Part I introduces the Bayesian approach and defines the information extraction as a two-level procedure: 1) model fitting, which is the incertitude alleviation over the model parameters, and 2) model selection, which is the incertitude alleviation over the class of models. The superiority of the Bayesian results is commented from an information theoretical perspective. The theoretical assay concludes with the proposal of a new systematic method for scene understanding from RS images: search for the scene that best explains the observed data. The method is demonstrated for high accuracy restoration of synthetic aperture radar (SAR) images with emphasis on new optimization algorithms for simultaneous model selection and parameter estimation. Examples are given for three families of Gibbs random fields (GRF) used as prior model libraries. Based on the Bayesian approach, a new method for optimal joint scale and model selection is demonstrated. Examples are given using a nested family of GRFs utilized as prior models for information extraction with applications both to SAR and optical images.

For pt.I see ibid., p.1431-45 (1998). The authors present Gibbs-Markov random field (GMRF) models as a powerful and robust descriptor of spatial information in typical remote-sensing image data. This class of stochastic image models provides an intuitive description of the image data using parameters of an energy function. For the selection among several nested models and the fit of the model, the authors proceed in two steps of Bayesian inference. This procedure yields the most plausible model and its most likely parameters, which together describe the image content in an optimal way. Its additional application at multiple scales of the image enables the authors to capture all structures being present in complex remote-sensing images. The calculation of the evidences of various models applied to the resulting quasicontinuous image pyramid automatically detects such structures. The authors present examples for both synthetic aperture radar (SAR) and optical data.