The capability to generate realistic hyperspectral imagery plays a prominent role in applications to sensor and mission development as well as in the training of machine learning models. Yet, it is a challenging task due to the high dimensionality and complex spectral–spatial structure of the data. This paper proposes a novel unsupervised deep-learning framework for generating realistic hyperspectral imagery based on blind hyperspectral unmixing and denoising diffusion probabilistic models. First, the approach extracts both endmembers and abundance maps from hyperspectral data through a dictionary of hyperspectral unmixing algorithms. The extracted abundances are then used as inputs for a guided diffusion model, which serves as the generative framework with the goal of producing realistic synthetic abundance maps. Finally, the generation of synthetic hyperspectral images is accomplished by integrating the generated abundance maps with the extracted endmember set and by suitably conditioning the probabilistic formulation of the guided diffusion model as a function of the unmixing algorithms in the aforementioned dictionary. By combining a collection of blind linear unmixing techniques with the generative capabilities of diffusion models, the proposed methodology aims to address key challenges in simulating hyperspectral sensor outputs. The methodology was validated experimentally using real satellite hyperspectral imagery from the PRISMA mission of the Italian Space Agency. The results confirm the effectiveness of the approach in generating realistic synthetic hyperspectral images associated with various land-covers. The code is available at: https://github.com/martinapastorino/HSI_DDPM.

Unsupervised change detection (CD) stands as a critical tool for damage assessment after a natural disaster. We emphasize heterogeneous CD methods, which support the case of highly heterogeneous images at the two observation dates, providing greater flexibility than traditional homogeneous methods. This adaptability is vital for swift responses in the aftermath of natural disasters. In this framework, we address the challenging case of detecting changes between the hyperspectral and synthetic aperture radar images. This case has intrinsic difficulties, namely, the difference in the nature of the physical quantity measured, added to the great difference in dimensionality of the two imaging domains. To address these challenges, a novel method is proposed based on the integration of a manifold learning technique and deep learning networks trained to perform an image-to-image translation task. The method works in a fully unsupervised manner, further enforcing a fast implementation in real-world scenarios. From an application-oriented perspective, we focus on flooded-area mapping using the PRISMA and COSMO-SkyMed missions. The experimental validation on two datasets, a semisimulated one and a real one associated with flooding, suggests that the proposed method allows for accurate detection of flooded areas and other ground changes.

Feature learning algorithms that use deep neural networks have shown to outperform traditional hand-crafted feature extraction methods when applied to satellite image time series. This learned features can be used in a multitude of applications such as classification, semantic segmentation, and change detection, among others. In this paper, we employ a feature learning technique to extract representative features from multimission polarimetric SAR (PolSAR) satellite image time series (SITS). To this end, we implement a formulation combining a 1-dimensional convolutional neural network and a stacked auto encoder. We performed experiments on a multimission and multitemporal PolSAR dataset and validated the extracted features through their utility as features in an unsupervised classification problem.

This paper tackles the semantic segmentation of zones affected by forest fires by the introduction of methods fusing multimodal imagery collected from unmanned aerial vehicles (UAVs) and satellite platforms. The multiresolution fusion task is especially challenging in this case because the difference between the involved spatial resolutions is very large – a situation that is normally not addressed by traditional multiresolution schemes. Two novel multiresolution fusion approaches, based on Bayesian and probabilistic graphical fusion models and integrated with a deep fully convolutional network and with the expectation-maximization algorithm, are proposed. The application is to a real case of fire zone mapping and management in the area of Marseille, France.

This paper explores remote sensing solutions for water-related disaster risk management within the RETURN (multi-Risk sciEnce for resilienT commUnities undeR a changiNg climate) project, funded within the Italian branch of the Next Generation EU framework. These encompass flood risk, drought risk, and coastal flooding under environmental and climatic changes. Emphasizing the pivotal role of remote sensing image processing, the paper discusses its significance as an essential tool within the project. It reviews remote sensing approaches tailored for the specified subtopics and delves into planned works, highlighting methodological approaches and data considerations crucial for addressing the complexities of water-related risks in the context of a changing climate.

The availability of multimodal remotely sensed images calls for the development of methods capable to jointly exploit the information deriving from images acquired at different spatial resolutions, frequencies, and bands, taking advantage from their possible complementary features. This letter proposes to address this task in the case of multimission synthetic aperture radar (SAR) images, through a combination of fully convolutional networks (FCNs), hierarchical probabilistic graphical models (PGMs), and decision tree ensembles. The objective is to model the multimodal information collected at multiple spatial resolutions by distinct space missions with SAR payloads through the nonparametric formulation of FCNs and decision trees, and the spatial and multiresolution modeling capabilities of FCNs and hierarchical PGMs. The experimental validation is conducted with multimission SAR imagery acquired at X-, L-, and C-band, respectively, by COSMO-SkyMed, SAOCOM, and Sentinel-1 over Northern Italy. The results suggest the advantages of incorporating multifrequency radar acquisitions to reach accurate classification maps and the multimodal fusion capabilities of the proposed methodology.

This article presents a method for the automatic learning of the potentials of a stochastic model, in particular a conditional random field (CRF), in a non-parametric fashion. The proposed model is based on a neural architecture, in order to leverage the modeling capabilities of deep learning (DL) approaches to directly learn semantic and spatial information from the input data. Specifically, the methodology is based on fully convolutional networks and fully connected neural networks. The idea is to access the multiscale information intrinsically extracted in the intermediate layers of a fully convolutional network through the integration of fully connected neural networks at different scales, while favoring the interpretability of the hidden layers as posterior probabilities. The potentials of the CRF are learned through an additional convolutional layer, whose kernel models the local spatial information considered. The loss function is computed as a linear combination of cross-entropy losses, accounting for the multiscale and the spatial information. To evaluate the capabilities of the proposed approach for the semantic segmentation of remote sensing images, the experimental validation was conducted with the ISPRS 2-D semantic labeling challenge Vaihingen and Potsdam datasets and with the IEEE GRSS data fusion contest Zeebruges dataset. As the ground truths of these benchmark datasets are spatially exhaustive, they have been modified to approximate the spatially sparse ground truths common in real remote sensing applications. The results are significant, as the proposed approach obtains higher average classification accuracies than recent state-of-the-art techniques considered in this article. The code is available athttps://github.com/Ayana-Inria/CRFNet-RS.

Image translation with convolutional autoencoders has recently been used as an approach to multimodal change detection (CD) in bitemporal satellite images. A main challenge is the alignment of the code spaces by reducing the contribution of change pixels to the learning of the translation function. Many existing approaches train the networks by exploiting supervised information of the change areas, which, however, is not always available. We propose to extract relational pixel information captured by domain-specific affinity matrices at the input and use this to enforce alignment of the code spaces and reduce the impact of change pixels on the learning objective. A change prior is derived in an unsupervised fashion from pixel pair affinities that are comparable across domains. To achieve code space alignment, we enforce pixels with similar affinity relations in the input domains to be correlated also in code space. We demonstrate the utility of this procedure in combination with cycle consistency. The proposed approach is compared with the state-of-the-art machine learning and deep learning algorithms. Experiments conducted on four real and representative datasets show the effectiveness of our methodology.

Among the effects of the climate change we are experiencing, the increase in the frequency of extreme event occurrences is evident. In this context, numerous studies confirmed the link between extreme meteorological events and the lightning activity. The possibility of having short-term predictions of the intensity of lightning phenomena would allow the near-real time monitoring of the evolution of such events. This paper proposes a multidisciplinary approach aiming at developing a regression algorithm to nowcast the density of cloud-to-ground lightning strokes one hour in advance. The proposed algorithm is developed thanks to the possibility to operate jointly with remote sensing imagery, numerical weather prediction model outcomes, and historical information on lightning. The possible dependence between meteorological data and lightning is seeked using Gaussian process regression models. The results obtained suggest that the proposed model can estimate low numbers of strokes accurately, whereas larger numbers of strokes are underestimated. Nevertheless, their presence is correctly detected. This suggests the potential of the proposed method as a processing tool to support the management of weather-related hazards.

This paper proposes a novel approach for heterogeneous Change Detection (CD) between hyperspectral (HS) and polarimetric synthetic aperture radar (PolSAR) images. The method utilizes an image-to-image (I2I) translation approach with cyclic generative networks to transfer data between the two domains. In this paper, we extend previous work where each image is transformed into the other domain by combining deep learning architectures and a formulation based on affinity matrices, thus resulting in two pairs of homogeneous images. Homogeneous CD algorithms are applied to both pairs, and the results are merged to generate a single change map. However, the disparity in the number of bands between HS and SAR data presents a challenge. To overcome this, a new method is introduced that combines dimensionality reduction with I2I translation and affinity matrix in a fully unsupervised manner. Experimental evaluation and comparative analysis with respect to a homogeneous CD method are presented.

This paper addresses the challenges of supervised semantic segmentation using Polarimetric Synthetic Aperture Radar (PolSAR) data for land cover mapping. We extend previous approaches relying on spatial-contextual classifier based on Support Vector Machines (SVMs) and Markov Random Field (MRF) models. The kernel used in this work extends a previously presented complex formulation based on reproducing kernel Hilbert spaces (RKHS). In this paper, we present a symmetrized form of this complex kernel, integrating it with global energy minimization techniques, and show that it provides more accurate predictions. The proposed approach achieves competitive accuracy on benchmark datasets, comparable to those of deep learning algorithms. The method's advantage lies in its lower resource requirements, making it a promising alternative for PolSAR semantic segmentation.

The problem of the semantic segmentation of multimodal images is characterized by the challenge of jointly exploiting information deriving from images possibly acquired at different spatial resolutions, frequencies, and bands. This paper proposes to address this task in the case of multimission synthetic aperture radar (SAR) images, through a combination of fully convolutional networks (FCNs), hierarchical probabilistic graphical models (PGMs), and decision tree ensembles. The objective is to model the spatial and multiresolution information contained in multimodal remote sensing images collected by distinct space missions with SAR payloads. The experimental validation is conducted with COSMO-SkyMed and SAOCOM images over Northern Italy. The results show that the proposed methodology is capable to reach accurate classification maps from input multimission SAR imagery.

Image segmentation is the process of partitioning a digital image into a set of homogeneous regions (according to some homogeneity criterion) to facilitate a subsequent higher-level analysis. In this context, the present paper proposes an unsupervised and graph-based method of image segmentation, which is driven by an application goal, namely, the generation of image segments associated with a user-defined and application-specific goal. A graph, together with a random grid of source elements, is defined on top of the input image. From each source satisfying a goal-driven predicate, called seed, a propagation algorithm assigns a cost to each pixel on the basis of similarity and topological connectivity, measuring the degree of association with the reference seed. Then, the set of most significant regions is automatically extracted and used to estimate a statistical model for each region. Finally, the segmentation problem is expressed in a Bayesian framework in terms of probabilistic Markov random field (MRF) graphical modeling. An ad hoc energy function is defined based on parametric models, a seed-specific spatial feature, a background-specific potential, and local-contextual information. This energy function is minimized through graph cuts and, more specifically, the alpha-beta swap algorithm, yielding the final goal-driven segmentation based on the maximum a posteriori (MAP) decision rule. The proposed method does not require deep a priori knowledge (e.g., labelled datasets), as it only requires the choice of a goal-driven predicate and a suited parametric model for the data. In the experimental validation with both magnetic resonance (MR) and synthetic aperture radar (SAR) images, the method demonstrates robustness, versatility, and applicability to different domains, thus allowing for further analyses guided by the generated products.

This paper presents a novel semantic segmentation method of very high resolution remotely sensed images based on fully convolutional networks (FCNs) and feedforward neural networks (FFNNs). The proposed model aims to exploit the intrinsic multiscale information extracted at different convolutional blocks in an FCN by the integration of FFNNs, thus incorporating information at different scales. The purpose is to obtain accurate classification results with realistic data sets characterized by sparse ground truth (GT) data by taking benefit from multiscale and long-range spatial information. The final loss function is computed as a linear combination of the weighted cross-entropy losses of the FFNNs and of the FCN. The modeling of spatial-contextual information is further addressed by the introduction of an additional loss term which allows to integrate spatial information between neighboring pixels. The experimental validation is conducted with the ISPRS 2D Semantic Labeling Challenge data set over the city of Vaihingen, Germany. The results are promising, as the proposed approach obtains higher average classification results than the state-of-the-art techniques considered, especially in the case of scarce, suboptimal GTs.

Decision fusion allows making a common decision by combining multiple opinions. In the context of remote sensing classification, such techniques are of great importance in all the cases where data collected by multiple sensors are merged into a final decision. Decision fusion may be used to combine the posterior probabilities associated with the output of single classifiers when applied to single sensor data. Meanwhile, techniques such as Markov Random Fields (MRFs) can integrate contextual information in the fusion process and are commonly used in classification. However, in the context of very large scale mapping (e.g., for global climate change monitoring), computation time can be critical and the application of both data fusion and spatial-contextual modeling comes with several constraints. In this paper, we propose a Bayesian decision fusion approach for optical-SAR image classification, integrated with a fast formulation of the iterated conditional modes (ICM) MRF-optimization algorithm based on a convolution operation. he validation on wide areas of Siberia proved the scalability and efficiency of the method for large scale applications.

Scene classification of remote sensing images is a challenging task due to the complexity and variety of natural scenes. In recent years, Convolutional Neural Networks (CNNs) have achieved impressive performances in many remote sensing scene classification benchmarks. However, in CNNs the long-range visual dependencies are often neglected due to the local filter design, leading to suboptimal performances in cluttered scenes such as urban areas. Recently proposed Transformer architecture resolved this issue by taking a broader neighborhood into account through the multi-head self-attention component. In this paper, we propose a novel method which borrows ideas from “knowledge distillation” and applied to recent vision Transformers. Specifically, we propose a compound loss computed on a Transformer-based student and a CNN teacher in a joint fashion and utilize it for the task of single-label scene classification. Because of the student's capability in capturing long-range visual dependencies, along with the inductive bias inherited from the teacher, our proposed model improves the classification accuracy on four well-known datasets compared to state-of-the-art approaches.

This paper addresses the semantic segmentation of synthetic aperture radar (SAR) images through the combination of ful-ly convolutional networks (FCN s), hierarchical probabilistic graphical models (PGMs), and decision tree ensembles. The idea is to incorporate long-range spatial information together with the multiresolution information extracted by FCN s, through the multiresolution graph topology on which hierar-chical PGMs can be efficiently formulated. The objective is to obtain accurate classification results with small datasets and reduce problems of spatial inconsistency. The experimental validation is conducted with several COSMO-SkyMed satel-lite images over Northern Italy. The results are significant, as the proposed method obtains more accurate classification results than the standard FCN s considered.

The automatic registration of image pairs composed of optical and synthetic aperture radar (SAR) images is a highly challenging task because of the inherently different physical, statistical, and textural properties of the input data. Information-theoretic measures capable of comparing local intensity distributions are often used for multisensor optical-SAR registration. Moreover, the growing availability of such heterogeneous data from current space missions require multisensor registration methods able to run on large-scale datasets with acceptable computation times. In this paper, a novel method is proposed combining information-theoretic area-based registration with a sequential image tiling strategy. Experiments with optical-SAR data collected by a variety of sensors (Sentinel, Landsat, ERS, etc.) suggest both qualitatively and quantitatively the effectiveness of the proposed strategy in achieving accurate registration with low computational cost.

Image translation with convolutional neural networks has recently been used as an approach to multimodal change detection. Existing approaches train the networks by exploiting supervised information of the change areas, which, however, is not always available. A main challenge in the unsupervised problem setting is to avoid that change pixels affect the learning of the translation function. We propose two new network architectures trained with loss functions weighted by priors that reduce the impact of change pixels on the learning objective. The change prior is derived in an unsupervised fashion from relational pixel information captured by domain-specific affinity matrices. Specifically, we use the vertex degrees associated with an absolute affinity difference matrix and demonstrate their utility in combination with cycle consistency and adversarial training. The proposed neural networks are compared with the state-of-the-art algorithms. Experiments conducted on three real data sets show the effectiveness of our methodology.

Convolutional neural networks (CNNs) represent the new reference approach for semantic segmentation of very-high-resolution (VHR) images, due to their ability to automatically capture semantic information while learning relevant features. However, as for most supervised methods, the map accuracy depends on the quantity and quality of ground truth (GT) used to train them. The use of densely annotated data (i.e., a detailed, exhaustive, pixel-level GT) allows to obtain effective CNN models but normally implies high efforts in annotation. Such ground truth is often available in benchmark datasets on which new methods are tested, but not on real data for land-cover applications, where only sparse annotations might be sufficiently cost effective. A CNN model trained with such incomplete GT maps has the tendency to smooth object boundaries because they are never precisely delineated in the GT. To cope with those shortcomings, we propose to exploit the intermediate activation maps of the CNN and to deploy a semisupervised fully connected conditional random field (CRF). In comparison with competitors using the same sparse annotations, the proposed method is able to better fill part of the performance gap compared to a CNN trained on the densely annotated, but generally unavailable, GTs.

Deep learning (DL) is currently the dominant approach to image classification and segmentation, but the performances of DL methods are remarkably influenced by the quantity and quality of the ground truth (GT) used for training. In this article, a DL method is presented to deal with the semantic segmentation of very-high-resolution (VHR) remote-sensing data in the case of scarce GT. The main idea is to combine a specific type of deep convolutional neural networks (CNNs), namely fully convolutional networks (FCNs), with probabilistic graphical models (PGMs). Our method takes advantage of the intrinsic multiscale behavior of FCNs to deal with multiscale data representations and to connect them to a hierarchical Markov model (e.g., making use of a quadtree). As a consequence, the spatial information present in the data is better exploited, allowing a reduced sensitivity to GT incompleteness to be obtained. The marginal posterior mode (MPM) criterion is used for inference in the proposed framework. To assess the capabilities of the proposed method, the experimental validation is conducted with the ISPRS 2D Semantic Labeling Challenge datasets on the cities of Vaihingen and Potsdam, with some modifications to simulate the spatially sparse GTs that are common in real remote-sensing applications. The results are quite significant, as the proposed approach exhibits a higher producer accuracy than the standard FCNs considered and especially mitigates the impact of scarce GTs on minority classes and small spatial details.

Road information extraction based purely on remote sensing can be affected by occlusions of the road surface caused by trees, shadows, and buildings. We propose a multimodal fusion method that addresses road extraction and road width estimation by combining aerial imagery, monocular images taken at ground level (street-level), and geospatial data (Open-StreetMap). The method combines semantic segmentation through convolutional neural networks, Voronoi diagram processing, and graph matching.

In this paper, a novel method to deal with the semantic segmentation of very high resolution remote sensing data is presented. Recent advances in deep learning (DL), especially convolutional neural networks (CNNs) and fully convolutional networks (FCNs), have shown outstanding performances in this task. However, the map accuracy depends on the quantity and quality of ground truth (GT) used to train them. At the same time, probabilistic graphical models (PGMs) have sparked even more interest in the past few years, because of the ever-growing need for structured predictions. The novel method proposed in this paper combines DL and PGMs to perform remote sensing image classification. FCNs can be exploited to deal with multiscale data through the integration with a hierarchical Markov model. The marginal posterior mode (MPM) criterion for inference is used in the proposed framework. Experimental validation is conducted on the ISPRS 2D Semantic Labeling Challenge Vaihingen dataset. The results are significant, as the proposed method has a higher recall than the standard FCNs considered and allows mitigating the impact of incomplete or suboptimal GT, especially with regard to the discrimination of minoritary classes.

Synthetic aperture radar (SAR) and optical satellite image registration is a field that developed in the last decades and gave rise to a great number of approaches. The registration process is composed of several steps: feature definition, feature comparison and optimization of a geometric transformation between the images. Feature definition can be done using simple traditional filtering or more complex deep learning (DL) methods. In this paper, two traditional approaches and a DL approach are compared. One can then wonder if the complexity of DL is worth to address the registration task. The aim of this paper is to quantitatively compare approaches rooted in distinct methodological areas on two common datasets with different resolutions. The comparison suggests that, although more complex, the DL approach is more precise than traditional methods.

The automatic registration of multisensor remote sensing images is a highly challenging task due to the inherently different physical, statistical, and textural properties of the input data. In the present paper, this problem is addressed in the case of optical-SAR images by proposing a novel method based on deep learning and area-based registration concepts. The method integrates a conditional generative adversarial network (cGAN), an area-based cross-correlation-type l2similarity metric, and the COBYLA constrained maximization algorithm. Whereas correlation-type metrics are typically ineffective in the application to multisensor registration, the proposed approach allows exploiting the image translation capabilities of cGAN architectures to enable the use of an l2similarity metric, which favors high computational efficiency. Experiments with Sentinel-1 and Sentinel-2 data suggest the effectiveness of this strategy and the capability of the proposed method to achieve accurate registration.

Change detection represents a major family of remote sensing image analysis techniques and plays a fundamental role in a variety of applications to environmental monitoring and disaster risk management. However, most change detection methods operate under the assumption that the multitemporal input data have been collected with the same (or very similar) acquisition modality - a possibly critical restriction in several applications. In this paper, the problem and the opportunities of change detection from multitemporal data acquired through heterogeneous modalities are addressed. Methodologically, this is a highly challenging data fusion problem, especially within an unsupervised framework. Here, these challenges and the methodological approaches proposed in the literature' which range from earlier semi-parametric regression to current deep learning architectures, are reviewed. Then, recent fully unsupervised techniques, based on spectral clustering, traditional image regression, and deep image-to-image translation, are briefly described.

In the framework of land-use mapping in urban areas, this paper explores the potential of the fusion of remote sensing data with information from transport demand data. The role of transport demand data is discussed and a probabilistic fusion framework is developed to exploit remote sensing and transport data in the discrimination of land use classes and land cover classes in urban and surrounding areas. Within this framework, two methods are proposed, based on pixelwise decision fusion and on the combination with a region-based multiscale Markov random field. The methods are validated on a case study associated with the Italian city of Genoa.

This paper proposes a new method for bitemporal change detection in heterogeneous remote sensing images. A modified canonical correlation analysis is used to align the code layers of two deep convolutional autoencoders, one for each image domain. It weights the input with a new affinity-based prior, which measures changes in pixel relations across the image domains and is used to reduce the influence of data points prone to change. By this procedure of self-supervision, we adapt the intrinsically supervised architecture to the unsupervised case, noting that the censoring of change pixels is key to efficiently learning the required data transformations. The result is an unsupervised algorithm which allows change detection in either of the image domains, or a combination of those, since efficient domain translation is obtained by coupling cross-domain encoders and decoders. We demonstrate state-of-the-art performance on real test datasets.

Because of the large variety of planetary sensors and spacecraft already collecting data and with many new and improved sensors being planned for future missions, planetary science needs to integrate numerous multimodal image sources, and, as a consequence, accurate and robust registration algorithms are required. In this article, we develop a new framework for crater detection based on marked point processes (MPPs) that can be used for planetary image registration. MPPs were found to be effective for various object detection tasks in Earth observation, and a new MPP model is proposed here for detecting craters in planetary data. The resulting spatial features are exploited for registration, together with fitness functions based on the MPP energy, on the mean directed Hausdorff distance, and on the mutual information. Two different methods—one based on birth–death processes and region-of-interest analysis and the other based on graph cuts and decimated wavelets—are developed within the proposed framework. Experiments with a large set of images, including 13 thermal infrared and visible images of the Mars surface, 20 semisimulated multitemporal pairs of images of the Mars surface, and a real multitemporal image pair of the Lunar surface, demonstrate the effectiveness of the proposed framework in terms of crater detection performance as well as for subpixel registration accuracy.

In this paper, we address the problem of the joint classification of multiple images acquired on the same scene at different spatial resolutions. From an application viewpoint, this problem is of importance in several contexts, including, most remarkably, satellite and aerial imagery. From a methodological perspective, we use a probabilistic graphical approach and adopt a hierarchical Markov mesh framework that we have recently developed and models the spatial-contextual classification of multiresolution and possibly multisensor images. Here, we focus on the methodological properties of this framework. First, we prove the causality of the model, a highly desirable property with respect to the computational cost of the inference. Then, we prove the expression of the marginal posterior mode criterion for this model and discuss the related assumptions. Experimental results with multi-spectral and panchromatic satellite images are also presented.

The monitoring of the recovery phase in the aftermath of an emergency scenario is tackled in this paper in terms of a change-detection perspective and through the integration of multisensor, multisource, and contextual information associated with high resolution optical and SAR data. The method makes use of the Markov random field theory to integrate the spatial context and the temporal correlation associated with images acquired at different dates. Moreover, the adoption of a region-based approach allows the characterization of the geometrical structures in the images through the employment of multiple segmentation maps at different scales. The performances of the proposed approach are evaluated on pairs of COSMO-SkyMed/Pléiades images acquired over Haiti in the aftermath of Hurricane Matthew.

In this paper, the problem of the classification of multiresolution and multisensor remotely sensed data is addressed by proposing a multiscale Markov mesh model. Multiresolution and multisensor fusion are jointly achieved through an explicitly hierarchical probabilistic graphical classifier, which uses a quadtree structure to model the interactions across different spatial resolutions, and a symmetric Markov mesh random field to deal with contextual information at each scale and favor applicability to very high resolution imagery. Differently from previous hierarchical Markovian approaches, here, data collected by distinct sensors are fused through either the graph topology itself (across its layers) or decision tree ensemble methods (within each layer). The proposed model allows taking benefit of strong analytical properties, most remarkably causality, which make it possible to apply time-efficient non-iterative inference algorithms.

Current satellite missions (e.g., COSMO-SkyMed, Sentinel-1) collect single- or multipolarimetric synthetic aperture radar (SAR) images with multiple spatial resolutions and possibly short revisit times. The availability of heterogeneous data requires effective methods able to exploit all the available information. In the context of environmental monitoring and natural disaster recovery, this paper proposes an unsupervised change detection method able to properly fuse and exploit multiresolution and multimodality SAR data. The data fusion process is based on the estimation of the virtual images that would have been collected in case all the sensors worked at the same spatial resolution and on the definition of a probabilistic model based on generalized Gaussian distributions and Gram-Charlier approximations. The detection of changes is addressed in a probabilistic graphical framework through a novel conditional random field, by defining an energy function that is minimized through graph-cuts or belief propagation methods.

Change detection (CD) in heterogeneous multitemporal satellite images is an emerging and challenging topic in remote sensing. In particular, one of the main challenges is to tackle the problem in an unsupervised manner. In this paper, we propose an unsupervised framework for bitemporal heterogeneous CD based on the comparison of affinity matrices and image regression. First, our method quantifies the similarity of affinity matrices computed from colocated image patches in the two images. This is done to automatically identify pixels that are likely to be unchanged. With the identified pixels as pseudotraining data, we learn a transformation to map the first image to the domain of the other image and vice versa. Four regression methods are selected to carry out the transformation: Gaussian process regression, support vector regression, random forest regression (RFR), and a recently proposed kernel regression method called homogeneous pixel transformation. To evaluate the potentials and limitations of our framework and also the benefits and disadvantages of each regression method, we perform experiments on two real data sets. The results indicate that the comparison of the affinity matrices can already be considered a CD method by itself. However, image regression is shown to improve the results obtained by the previous step alone and produces accurate CD maps despite of the heterogeneity of the multitemporal input data. Notably, the RFR approach excels by achieving similar accuracy as the other methods, but with a significantly lower computational cost and with fast and robust tuning of hyperparameters.

In this paper, the problem of the spatial-spectral classification of very high-resolution optical images is addressed using a kernel- and region-based approach. A novel method based on integrating region-based or object-based information into a kernel machine is developed. A Gaussian process model is used to characterize each segment in a segmentation map and to define a region-based admissible kernel accordingly. This kernel is combined with a marker-controlled watershed segmentation that incorporates scale adaptivity. Spatial-spectral fusion capabilities are also ensured by combining the resulting classification method with composite kernels.

Convolutional Neural Networks (CNNs) have become the new standard for semantic segmentation of very high resolution images. But as for other methods, the map accuracy depends on the quantity and quality of ground truth used to train them. Having densely annotated data, i.e. a detailed, pixel-level ground truth (GT), allows obtaining effective models, but requires high efforts in annotation. For this reason, it is more common and efficient to work with point or scribbled annotations rather than with dense ones. A CNN model trained with such incomplete ground truths tends to mischaracterize the shapes of the objects and to be inaccurate near their boundaries. We propose to use an approximation of a fully connected Conditional Random Field (CRF) to solve these issues, in which long range connections are accounted for through auxiliary nodes based on clustering of CNN activation features. Experiments on the ISPRS Vaihingen benchmark, where a CNN is trained only with a non-dense, scribbled ground truth, show that the proposed method can fill part of the performance gap with respect to models trained on the densely annotated, but unrealistic, ground truth.

Classification of remotely sensed images into land cover or land use is highly dependent on geographical information at least at two levels. First, land cover classes are observed in a spatially smooth domain separated by sharp region boundaries. Second, land classes and observation scale are also tightly intertwined: they tend to be consistent within areas of homogeneous appearance, or regions, in the sense that all pixels within a roof should be classified as roof, independently on the spatial support used for the classification. In this paper, we follow these two observations and encode them as priors in an energy minimization framework based on conditional random fields (CRFs), where classification results obtained at pixel and region levels are probabilistically fused. The aim is to enforce the final maps to be consistent not only in their own spatial supports (pixel and region) but also across supports, i.e., by getting the predictions on the pixel lattice and on the set of regions to agree. To this end, we define an energy function with three terms: 1) a data term for the individual elements in each support (support-specific nodes); 2) spatial regularization terms in a neighborhood for each of the supports (support-specific edges); and 3) a regularization term between individual pixels and the region containing each of them (intersupports edges). We utilize these priors in a unified energy minimization problem that can be optimized by standard solvers. The proposed 2LCRF model consists of a CRF defined over a bipartite graph, i.e., two interconnected layers within a single graph accounting for interlattice connections. 2LCRF is tested on two very high-resolution data sets involving submetric satellite and subdecimeter aerial data. In all cases, 2LCRF improves the result obtained by the independent base model (either random forests or convolutional neural networks) and by standard CRF models enforcing smoothness in the spatial domain.

In order to ensure homogeneity in performance assessment of proposed algorithms for information extraction in the Earth Observation (EO) domain, standardized remotely sensed datasets are particularly useful and welcome. Fully aware of this principle, the IEEE Geoscience and Remote Sensing Society (GRSS) and especially its Image Analysis and Data Fusion Technical Committee (IADF), has been organizing for some years now the Data Fusion Contest (DFC). In the DFC, one specific dataset is made available to the scientific community, which can download it and use it to test its newly developed algorithms. The consistence of the starting dataset across participating groups ensures the significance of assessing and ranking results, to finally proclaim the winner who scored the highest. More recently, the IEEE GRSS has provided one more contribution to the standardization effort by building the Data and Algorithm Standard Evaluation (DASE) website. DASE can distribute to registered users a limited set of possible “standard” open datasets, together with some ground truth info, and automatically assess the processing results provided by the users. In this paper we report on the birth of this initiative and present some recently introduced features.

This paper addresses the problem of exploiting very high-resolution multifrequency SAR data collected by the COSMO-SkyMed and RADARSAT-2 missions to support risk monitoring and assessment in urban and suburban areas. The proposed approach aims at taking benefit from the synergy between the two SAR data sources to optimize the accuracy of thematic products of interest to risk monitoring. In particular, an unsupervised change detection approach is discussed, in which feature-level fusion is applied to a satellite image time series including both COSMO-SkyMed and RADARSAT-2 acquisitions to improve the detection results as compared to those generated through single-frequency processing.

This paper addresses the problem of semi-automatic image registration on planetary images. A joint feature-based and area-based approach is proposed. Firstly, the most relevant craters are extracted from the two images to register, and then, registration is performed in two steps. The first step matches the craters extracted from the images based on a generalized Hausdorff distance. In the second step, the mutual information between the two images is maximized to achieve high registration accuracy. Craters are detected by a stochastic-geometry approach based on a marked point process model and of a multiple-birth-and-cut energy minimization algorithm. The experimental validation is carried out with 13 images for the crater extraction stage, and with 20 semi-synthetic pairs of images with ground truth and several images extracted from actual multitemporal lunar scenes for the registration phase.

Because of the large variety of sensors and spacecraft collecting data, planetary science needs to integrate various multisensor and multitemporal images. These multiple data represent a precious asset, as they allow the study of target spectral responses and of changes in the surface structure. Because of their variety, they also require accurate and robust registration. A new crater detection algorithm, used to extract features to be integrated in an image registration framework, is presented. A marked point process-based method has been developed to model the spatial distribution of elliptical objects (i.e. the craters), and a birth-death Markov chain method, coupled with a region-based scheme aiming at computational efficiency, is used to find the optimal configuration fitting the image. The extracted features are exploited, together with a newly defined fitness function based on a modified Hausdorff distance, by an image registration algorithm whose architecture has been designed to minimize the computation time.

This letter proposes two methods for the supervised classification of multisensor optical and synthetic aperture radar images with possibly different spatial resolutions. Both the methods are formulated within a unique framework based on hierarchical Markov random fields. Distinct quad-trees associated with the individual information sources are defined to jointly address multisensor, multiresolution, and possibly multifrequency fusion, and are integrated with finite mixture models and the marginal posterior mode criterion. Experimental validation is conducted with Pléiades, COSMO-SkyMed, RADARSAT-2, and GeoEye-1 data.

In this paper, we propose a novel hierarchical method for remote sensing image classification. The proposed approach integrates an explicit hierarchical graph-based classifier, which uses a quad-tree structure to model multiscale interactions, and a third order Markov mesh random field to deal with pixel wise contextual information in the same scale. The choice of a quad-tree and the third order Markov mesh allow taking benefit from their good analytical properties (especially causality) and consequently apply non-iterative algorithms. Indeed, the Markov mesh is used to incorporate spatial information in each scale of the quad-tree while keeping the causality of the hierarchical model.

Hyperspectral images in the thermal infrared range are attracting increasing attention in the remote sensing field. Nonetheless, the generation of land cover maps using this innovative kind of remote sensing data has been scarcely studied so far. The aim of this article is to experimentally investigate the potential of various supervised classification approaches to land cover mapping from high spatial resolution thermal hyperspectral images. The considered methods include both non-contextual and spatial-contextual classifiers, and encompass methodological approaches based on Bayesian decision theory, Markov random fields, multiscale region-based analysis, and Bayesian feature reduction. Experiments were conducted with a challenging data set associated with a complex urban and vegetated scene. Overall accurate results were achieved by using contextual approaches. The validation suggested the effectiveness of pattern recognition tools in the application to this innovative typology of remote sensing data while also indicating potential improvements through the fusion with physically-based methods.

The availability of synthetic aperture radar (SAR) data with high spatial resolution offers great potential for environmental monitoring due to the insensitivity of SAR to atmospheric and sunlight-illumination conditions. In this paper, an unsupervised change detection method for SAR images at medium to high resolution is proposed. The image ratioing approach is adopted, and a Bayesian unsupervised minimum-error thresholding algorithm is extended by proposing a technique based on Generalized Gamma distributions (GΓD). GΓD was recently found to be an accurate model for the statistics of SAR amplitudes at moderate to high resolution. Here, a specific parametric modeling approach for the ratio of G D-distributed SAR images is proposed and endowed with a probability density function estimation algorithm based on the method of log-cumulants. Consistency of this estimator is proven. Experimental results confirm the accuracy of the method for medium and high resolutions X-band SAR images.

In this paper we address the problem of urban optical imagery classification by developing a convolutional neural network (CNN) approach. We design a custom CNN that operates on local patches in order to produce dense pixel-level classification map. In this work we focus on a comprehensive dataset of 2.5-meter SPOT-5 imagery acquired at different dates and sites. The performance of the proposed model is validated on a five target-class problem and compared with a benchmark random forest classifier with a set of hand-picked features.

Land cover / land use classification of remotely sensed images is inherently geographical. The use of spatial information, accounting for neighborhood relationship and spatial smoothness of geographical objects, made its proofs in countless occasions and, especially when considering very high resolution images, methods ignoring spatial context do not perform well. In this paper, we propose a hybrid dual-layer conditional random field model that enforces spatial smoothness and consistency between the pixel and region-based maps. We formulate these intuitions as a standard energy minimization problem, and we show that finding a joint solution over both output spaces leads to strong improvements in the numerical and visual senses.

In this paper, we propose a new test statistic for unsupervised change detection in polarimetric radar images. We work with multilook complex covariance matrix data, whose underlying model is assumed to be the scaled complex Wishart distribution. We use the complex-kind Hotelling-Lawley trace (HLT) statistic for measuring the similarity of two covariance matrices. The distribution of the HLT statistic is approximated by a Fisher-Snedecor distribution, which is used to define the significance level of a false alarm rate regulated change detector. Experiments on simulated and real PolSAR data sets demonstrate that the proposed change detection method gives detection rates and error rates that are comparable with the generalized likelihood ratio test.

In this paper, we propose a novel method for the joint classification of both multidate and multiresolution remote sensing imagery, which represents an important and relatively unexplored classification problem. The proposed classifier is based on an explicit hierarchical graph-based model that is sufficiently flexible to address a coregistered time series of images collected at different spatial resolutions. Within this framework, a novel element of the proposed approach is the use of multiple quadtrees in cascade, each associated with the images available at each observation date in the considered time series. For each date, the input images are inserted in a hierarchical structure on the basis of their resolutions, whereas missing levels are filled in with wavelet transforms of the images embedded in finer-resolution levels. This approach is aimed at both exploiting multiscale information, which is known to play a crucial role in high-resolution image analysis, and supporting input images acquired at different resolutions in the input time series. The experimental results are shown for multitemporal and multiresolution optical data.

The problem of supervised classification of multiresolution images, which are composed of a higher resolution panchromatic channel and of several coarser resolution multispectral channels, is addressed in this paper by proposing a novel contextual method based on Markov random fields. The method iteratively exploits a linear mixture model for the relationships between data at different resolutions and a graph cut approach to Markovian energy minimization to generate a contextual classification map at the highest resolution available in the input data set. The estimation of the parameters of the method is performed by extending recently proposed techniques based on the expectation-maximization and Ho-Kashyap's algorithms. The method is experimentally validated with semisimulated and real data involving both IKONOS and Landsat-7 ETM+ images, and the results are compared with those generated by previous approaches to the classification of multiresolution imagery.

In this paper, we address the problem of unsupervised change detection on two or more coregistered images of the same object or scene at several time instants. We propose a novel empirical-Bayesian approach that is based on a false discovery rate formulation for statistical inference on local patch-based samples. This alternative error metric allows to efficiently adjust the family-wise error rate in case of the considered large-scale testing problem. The designed change detector operates in an unsupervised manner under the assumption of the limited amount of changes in the analyzed imagery. The detection is based on the use of various statistical features, which enable the detector to address application-specific detection problems provided an appropriate ad hoc feature choice. In particular, we demonstrate the use of the rank-based statistics: Wilcoxon and Cramér-von Mises for image pairs, and multisample Levene statistic for short image sequences. The experiments with remotely sensed radar, dermatological, and still camera surveillance imagery demonstrate accurate performance and flexibility of the proposed method.

This paper addresses the problem of parameter optimization for Markov random field (MRF) models for supervised classification of remote sensing images. MRF model parameters generally impact on classification accuracy, and their automatic optimization is still an open issue especially in the supervised case. The proposed approach combines a mean square error (MSE) formulation with Platt's sequential minimal optimization algorithm, with the aim of taking benefit from the effectiveness of this quadratic programming technique in both computation time and memory occupation. The experimental validation is carried out with five real data sets comprising multipolarization and multifrequency SAR, multispectral high-resolution, single date and multitemporal imagery. The method is compared with two techniques based on MSE criteria and on the Ho-Kashyap and Goldfard-Idnani numerical algorithms.

This paper addresses the problem of multisensor fusion of COSMO-SkyMed and RADARSAT-2 data together with optical imagery for classification purposes. The proposed method is based on an explicit hierarchical graph-based model that is sufficiently flexible to deal with multisource coregistered images collected at different spatial resolutions by different sensors. An especially novel element of the proposed approach is the use of multiple quad-trees in cascade, each associated with a set of images acquired by different SAR sensors, with the aim to characterize the correlations associated with distinct images from different instruments. Experimental results are shown with COSMO-SkyMed, RADARSAT-2, and Pléiades data1.

A heuristic utilizing both spectral and spatial information is proposed for active learning. It addresses the issue of iteratively querying most informative training samples with a special focus on spatial-contextual image classification. With the aim to utilize all information during the learning process, the proposed heuristic queries unlabeled pixels considering spectral-spatial inconsistency (SSI), i.e., the unlabeled pixels whose spectral and spatial information indicate different class labels are favored in the active selection. To model spectral-spatial information, a Markov random field (MRF), in which the unary term is defined using the output of a support vector machine and the pairwise term is defined by a multilevel logistic model, is adopted. A new approach to the estimation of the parameters of this MRF model is also incorporated in the proposed method. It aims at taking benefit of spatial information by using the pixels which are representative of the inter-class spatial transitions. A high resolution remotely sensed image is used in the experiments, and the proposed method is proved to be feasible and accurate.

Earth observation through remote sensing images allows the accurate characterization and identification of materials on the surface from space and airborne platforms. Multiple and heterogeneous image sources can be available for the same geographical region: multispectral, hyperspectral, radar, multitemporal, and multiangular images can today be acquired over a given scene. These sources can be combined/fused to improve classification of the materials on the surface. Even if this type of systems is generally accurate, the field is about to face new challenges: the upcoming constellations of satellite sensors will acquire large amounts of images of different spatial, spectral, angular, and temporal resolutions. In this scenario, multimodal image fusion stands out as the appropriate framework to address these problems. In this paper, we provide a taxonomical view of the field and review the current methodologies for multimodal classification of remote sensing images. We also highlight the most recent advances, which exploit synergies with machine learning and signal processing: sparse methods, kernel-based fusion, Markov modeling, and manifold alignment. Then, we illustrate the different approaches in seven challenging remote sensing applications: 1) multiresolution fusion for multispectral image classification; 2) image downscaling as a form of multitemporal image fusion and multidimensional interpolation among sensors of different spatial, spectral, and temporal resolutions; 3) multiangular image classification; 4) multisensor image fusion exploiting physically-based feature extractions; 5) multitemporal image classification of land covers in incomplete, inconsistent, and vague image sources; 6) spatiospectral multisensor fusion of optical and radar images for change detection; and 7) cross-sensor adaptation of classifiers. The adoption of these techniques in operational settings will help to monitor our planet from space in the very near future.

In this paper, we propose a novel method for the joint classification of multidate, multiresolution and multisensor remote sensing imagery, which represents a vital and fairly unexplored classification problem. The proposed classifier is based on an explicit hierarchical graph-based model sufficiently flexible to deal with multisource coregistered time series of images collected at different spatial resolutions [1]. An especially novel element of the proposed approach is the use of multiple quadtrees in cascade, each associated with each new available image at different dates, with the aim to characterize the temporal correlations associated with distinct images in the input time series. Experimental results are shown with multitemporal and multiresolution Pléiades data.

The problem of remote-sensing image classification is addressed in this paper by proposing a novel contextual classification method that integrates support vector machines (SVMs), Markov random fields (MRFs), and graph cuts. The proposed approach is methodologically explained by the aim to combine the robustness to dimensionality issues and the generalization capability of SVMs, the effectiveness of Markov models in characterizing the spatial contextual information associated with an image, and the capability of graph cut techniques in tackling complex problems of global minimization in computationally acceptable times. In the proposed method, the MRF minimum-energy problem is formalized in terms of an appropriate SVM kernel expansion and addressed through graph cuts. Parameter estimation is automated through two specific algorithms, based on the Ho-Kashyap and Powell numerical procedures. Experiments are carried out with two data sets consisting of multichannel SAR and multispectral high-resolution images.

In this paper, we propose a novel method to fuse multidate, multiresolution, and multiband remote sensing imagery for multitemporal classification purposes. The proposed method is based on an explicit hierarchical graph-based model that is sufficiently flexible to deal with multisource coregistered time series of images collected at different spatial resolutions. An especially novel element of the proposed approach is the use of multiple quad-trees in cascade, each associated with an image acquired at a different date, with the aim to characterize the temporal correlations associated with distinct images in an input time series. Experimental results are shown with multitemporal and multiresolution Pléiades data1.

In the context of sea and ocean monitoring by Earth Observation (EO) data, this paper illustrates the activity of the SEAGOSS1(“Information, warning, and management system for sea water pollution from oil slick and sediments,” 2012-2014) project, carried out in collaboration with an Italian SME Gruppo SIGLA S.r.l. and DHI Italia (former Danish Hydraulic Institute) both located in Genoa, Italy. The proposed paper presents here the added value of advanced pattern recognition techniques to process remote sensing data to model sea state, in application as coastal monitoring and oil slick detection.

Accurately mapping air temperature near the Earth surface plays a primary role in weather and climate studies and for solar energy planning and production. Remote sensing allows spatially distributed estimates of air temperature to be computed, thus complementing the spatially sparse observations collected by ground stations. In this paper, a novel method for periodic (e.g., daily or monthly) air-temperature estimation from satellite images is proposed. It is based on support vector machines (SVMs) and generalizes, to the case of air temperature, a recently developed SVM-based approach to land and sea surface temperature estimation. Case-specific techniques aimed at computing periodic statistics of air temperature and based on the expectation-maximization algorithm are also integrated in the proposed approach. The method also allows the statistics of the estimation error to be modeled on a pixelwise basis by combining nonstationary stochastic processes and Clark's variance approximation. Experimental results with MSG-SEVIRI and MétéoFrance data acquired over Provence-Alpes-Côte d'Azur (France) are presented.

A kernel-based approach is proposed in this paper to address supervised classification of polarimetric SAR data. Relevant features extracted from such data are generally complex-valued (e.g., scattering coefficients, multilook covariance-matrix entries). First, based on the theory of complex reproducing kernel Hilbert spaces (RKHS's), a family of admissible kernel functions tailored to the classification of complex-valued features is proposed. Then, a support vector machine (SVM) classifier is developed using this family of kernels and a case-specific interpretation is discussed for the related notion of maximum-margin hyperplane in a complex vector space. Finally, a spatial-contextual classifier is introduced by integrating the proposed family of kernels with a recent combination of SVM and Markov random fields. Case-specific techniques, based on the Powell and Ho-Kashyap numerical algorithms, are incorporated in the proposed methods to automatically optimize their parameters. Experiments with SIR-C data are discussed.

The quantity and the updating time of the archives of very high spatial resolution visible and near-infrared remote sensing images for commercial use improved during the last years. This led to the detection of changes on the Earth surface through remote sensing images to become a key analytical tool for many public and private organizations, which can take advantage of the information carried out to help and improve their decision making processes. This paper proposes an unsupervised method for detecting multiple changes in the application to damage assessment after a flood. It is composed of five steps, and is based on a change vector analysis approach. After a case-specific feature extraction stage, through a process called normalized difference indexing, the change detection task is carried out by modeling the classes of changed and not changed pixels with a Gaussian finite mixture model, using the expectation-maximization algorithm to estimate the statistical parameters involved. Then, the mean shift clustering algorithm is used to discriminate among different types of change. The method has been tested on a pair of images acquired by WorldView-2 and associated with the 2013 flood in Colorado.

In this paper, we develop a novel classification approach for multiresolution, multisensor [optical and synthetic aperture radar (SAR)], and/or multiband images. This challenging image processing problem is of great importance for various remote sensing monitoring applications and has been scarcely addressed so far. To deal with this classification problem, we propose a two-step explicit statistical model. We first design a model for the multivariate joint class-conditional statistics of the coregistered input images at each resolution by resorting to multivariate copulas. Such copulas combine the class-conditional marginal probability density functions (pdfs) of each input channel that are estimated by finite mixtures of well-chosen parametric families. We consider different distribution families for the most common types of remote sensing imagery acquired by optical and SAR sensors. We then plug the estimated joint pdfs into a hierarchical Markovian model based on a quad-tree structure, where each tree-scale corresponds to the different input image resolutions and to corresponding multiscale decimated wavelet transforms, thus preventing a strong resampling of the initial images. To obtain the classification map, we resort to an exact estimator of the marginal posterior mode. We integrate a prior update in this model in order to improve the robustness of the developed classifier against noise and speckle. The resulting classification performance is illustrated on several remote sensing multiresolution data sets, including very high resolution and multisensor images acquired by COSMO-SkyMed and GeoEye-1.

In this paper we address the problem of unsupervised change detection on image pairs. We develop a novel patch-based hypothesis testing approach that employs the false discovery rate technique for statistical hypothesis testing. The designed approach can be adopted to specific detection applications via selection of appropriate statistical features. Experiments with still camera imagery demonstrate high performance and flexibility of the proposed method.

The problem of unsupervised change detection from multisensor very high resolution images is addressed in this paper by focusing on the case in which multitemporal SAR data but only a single-date optical observation are available. This peculiar and challenging scenario is especially interesting in disaster management applications in which SAR acquisitions are feasible both before and after the event and an optical image is available only at one date (e.g., from the archive). The proposed method combines a novel Markov random field model with multiscale region-based analysis in order to fuse the information associated both with the statistics of the ratio of the multitemporal SAR images and with the spatial-geometrical structure of the observed scene captured by the optical image. Parameter estimation is based on a dictionary of parametric families and is carried out through the expectation-maximization algorithm and the method of log-cumulants. Graph cuts are used to minimize the energy function of the proposed MRF model. Experimental results are presented with COSMO-SkyMed and GeoEye-1 images.

Satellite synthetic aperture radar (SAR) systems currently offer both very high resolutions and multiresolution acquisition capability, thus presenting a great potential for environmental monitoring and damage assessment applications. In this framework, change detection methods play a central role. In this paper, a novel unsupervised change detection method is proposed for multitemporal SAR images acquired at multiple resolutions. The method combines Markov random field modeling, line processes, linear mixtures, Bayesian estimation, generalized Gaussian distributions, and graph cuts with the aim of fusing the available multiresolution information to generate a change map at the finest of the observed resolutions. The proposed method is experimentally validated with multitemporal COSMO-SkyMed stripmap and polarimetric data.

This letter addresses the problem of classifying synthetic aperture radar (SAR) images of urban areas by using a supervised Bayesian classification method via a contextual hierarchical approach. We develop a bivariate copula-based statistical model that combines amplitude SAR data and textural information, which is then plugged into a hierarchical Markov random field model. The contribution of this letter is thus the development of a novel hierarchical classification approach that uses a quad-tree model based on wavelet decomposition and an innovative statistical model. The performance of the developed approach is illustrated on a high-resolution satellite SAR image of urban areas.

Markov models represent a wide and general family of stochastic models for the temporal and spatial dependence properties associated to 1-D and multidimensional random sequences or random fields. Their applications range over a wide variety of subareas of the information and communication technology (ICT) field, including networking, automation, speech processing, genomic-sequence analysis, or image processing. Focusing on the applicative problem of land-cover mapping from very-high-resolution (VHR) remote sensing images, which is a relevant problem in many applications of environmental monitoring and natural resource exploitation, Markov models convey a great potential, thanks to their capability to effectively describe and incorporate the spatial information associated with image data into an image-classification process. In this framework, the main ideas and previous work about Markov modeling for VHR image classification will be recalled in this paper and processing results obtained through recent methods proposed by the authors will be discussed.

This paper proposes a novel unsupervised, non-Gaussian, and contextual segmentation method that combines an advanced statistical distribution with spatial contextual information for multilook polarimetric synthetic aperture radar (PolSAR) data. This extends on previous studies that have shown the added value of both non-Gaussian modeling and contextual smoothing individually or for intensity channels only. The method is based on a Markov random field (MRF) model that integrates aK-Wishart distribution for the PolSAR data statistics conditioned to each image cluster and a Potts model for the spatial context. Specifically, the proposed algorithm is constructed based upon the stochastic expectation maximization (SEM) algorithm. A new formulation of SEM is developed to jointly perform clustering of the data and parameter estimation of theK-Wishart distribution and the MRF model. Experiments on simulated and real PolSAR data demonstrate the added value of using an appropriate statistical representation, in combination with contextual smoothing.

In the framework of remote-sensing image classification, support vector machines (SVMs) have lately been receiving substantial attention due to their accurate results in many applications as well as their remarkable generalization capability even with high-dimensional input data. However, SVM classifiers are intrinsically noncontextual, which represents an important limitation in image classification. In this paper, a novel and rigorous framework, which integrates SVMs and Markov random field models in a unique formulation for spatial contextual classification, is proposed. The developed contextual generalization of SVMs, is obtained by analytically relating the Markovian minimum-energy criterion to the application of an SVM in a suitably transformed space. Furthermore, as a second contribution, a novel contextual classifier is developed in the proposed general framework. Two specific algorithms, based on the Ho–Kashyap and Powell numerical procedures, are combined with this classifier to automate the estimation of its parameters. Experiments are carried out with hyperspectral, multichannel synthetic aperture radar, and multispectral high-resolution images and the behavior of the method as a function of the training-set size is assessed.

Parameter estimation of probability density functions is one of the major steps in the area of statistical image and signal processing. In this paper we explore several properties and limitations of the recently proposed method of logarithmic cumulants (MoLC) parameter estimation approach which is an alternative to the classical maximum likelihood (ML) and method of moments (MoM) approaches. We derive the general sufficient condition for a strong consistency of the MoLC estimates which represents an important asymptotic property of any statistical estimator. This result enables the demonstration of the strong consistency of MoLC estimates for a selection of widely used distribution families originating from (but not restricted to) synthetic aperture radar image processing. We then derive the analytical conditions of applicability of MoLC to samples for the distribution families in our selection. Finally, we conduct various synthetic and real data experiments to assess the comparative properties, applicability and small sample performance of MoLC notably for the generalized gamma and K families of distributions. Supervised image classification experiments are considered for medical ultrasound and remote-sensing SAR imagery. The obtained results suggest that MoLC is a feasible and computationally fast yet not universally applicable alternative to MoM. MoLC becomes especially useful when the direct ML approach turns out to be unfeasible.

This paper considers change detection with multitemporal synthetic aperture radar (SAR) images. A novel statistical approach is developed that formulates the detection of changes between two coregistered SAR images acquired at different dates as a non-parametric local-window hypothesis testing problem. This approach takes into account the simultaneous testing of a large amount of similar hypotheses (i.e., one Wilcoxon test at each pixel of the image), and is based on a likelihood ratio test to obtain a detection map at the pixel level. Some encouraging detection results are obtained on XSAR and very high resolution COSMO-SkyMed images.

Change detection techniques represent important tools for environmental monitoring and damage assessment after environmental disasters. However, change detection methods that were found accurate for coarser-resolution SAR are often ineffective with current very high resolution (VHR) satellite SAR due to the need to suitably model the contextual and geometrical information associated with VHR data. In this paper, a novel unsupervised change detection technique is proposed for VHR SAR based on Markov random fields (MRFs), line processes, and a dictionary of SAR-specific probability density models. The estimation of the parameters of the proposed MRF model is carried out through the expectation-maximization algorithm and the method of log-cumulants. Graph cuts are used to minimize the energy function of the MRF model because of their capability to approach global minima or strong local minima in acceptable computation times. The proposed method is experimented with COSMO-SkyMed images acquired before and after an earthquake.

In the framework of the monitoring of structures and infrastructures from environmental disasters, the COSMO-SkyMed constellation has a huge potential, thanks to up to metric spatial resolution, short revisit time, and the day/night all-weather acquisition capability ensured by SAR. This paper focuses on the scientific results of the project “Development and validation of multitemporal image analysis methodologies for multirisk monitoring of critical structures and infrastructures,” funded by the Italian Space Agency. Several change-detection, data-fusion, and feature-extraction techniques, which were developed and experimentally validated in the project for COSMO-SkyMed imagery and for their integration with other data sources (including very high resolution optical data), are described and examples of processing results are discussed.

With the launch of several planetary missions in the last decade, a large amount of planetary images has been already acquired and much more will be available for analysis in the coming years. The image data need to be analyzed, preferably by automatic processing techniques because of the huge amount of data. Although many automatic feature extraction methods have been proposed and utilized for earth remote sensing images, these methods are not always applicable to planetary data that often present low contrast and uneven illumination characteristics. Here, we propose a new unsupervised method for the extraction of different features of elliptical and geometrically compact shapes, such as craters and rocks of compact shape (e.g., boulders), to be used for image registration purposes. This approach is based on the combination of several image processing techniques, including watershed segmentation and the generalized Hough transform. The method potentially has application for extraction of craters, rocks, and other geological features.

Satellite remote sensing missions devoted to Earth observation (EO) currently offer a unique capability to monitor the evolution of the Earth's surface by providing temporally repetitive views at the desired (global, regional, or local) spatial scale. This wealth of remote sensing data conveys a huge potential for preventing, monitoring, and managing natural or man-made disasters. Specifically focusing on flood risk, a successful exploitation of this potential requires not only accurate and reliable image-analysis methods to extract the desired thematic information, but also the ability to combine this information with physically based models of the observed processes. Therefore, a multidisciplinary approach combining remote sensing with geophysical sciences, such as, in this case, hydrometeorology, is fundamental. This combination of expertise allows, in particular, satellite data to be exploited within the different phases of flood risk reduction: risk assessment, prevention, mitigation, monitoring, and management. In this paper, we investigate the key issues involved in the exploitation of satellite data with special focus on the phases of the emergency and post-disaster damage assessment. To this end, the challenges and the methodological approaches involved in the multidisciplinary combination of image analysis and hydrometeorology are discussed with the purpose of guiding and optimizing the process of information extraction from satellite data according to the requirements of civil protection from floods. Experimental examples of a few relevant case studies are also presented.

A clustering method that combines an advanced statistical distribution with spatial contextual information is proposed for multilook polarimetric synthetic aperture radar (PolSAR) data. It is based on a Markov random field (MRF) model that integrates a K-Wishart distribution for the PolSAR data statistics conditioned to each image cluster and a Potts model for the spatial context. Specifically, the proposed algorithm is constructed based upon the expectation maximization (EM) algorithm. A new formulation of EM is developed to jointly address parameter estimation in the K-Wishart distribution and the spatial context model, and also minimization of the energy function. Experiments are presented with simulated and real quad-pol L-band data.

The problem of joint classification of multitemporal high-resolution images is addressed in this paper by proposing a novel multiscale region-based technique. Given a pair of multitemporal images acquired over the same area, multiscale segmentation is applied to each image in order to generate a collection of segmentation results related to different spatial scales. A novel Markov random field (MRF) model is developed to fuse the resulting multiscale information together with the spatial and temporal contextual information associated with the input multitemporal data set. The parameters of the MRF model are automatically optimized by a recent technique based on the Ho-Kashyap's algorithm. Experiments are presented with QuickBird and SPOT-5 data.

Spatial-contextual classification methods based either on stochastic Markov random field (MRF) models, on texture analysis, or on region-based processing are important tools for high-resolution multispectral image analysis. In this paper, a novel supervised classification technique is proposed, that integrates the MRF, texture-based, and region-based approaches to contextual image classification in a unique multiscale framework. A previous method, based on the combination of MRFs with multiscale segmentation, is generalized and integrated with the multivariate semivariogram approach to texture analysis. In order to minimize the impact of texture-extraction artifacts at the spatial edges between different classes, an adaptive semivariogram-estimation technique is also developed and iteratively incorporated in the proposed classifier. Experiments are presented with IKONOS images.

In this letter, we address the problem of estimating the amplitude probability density function (pdf) of single-channel synthetic aperture radar (SAR) images. A novel flexible method is developed to solve this problem, extending the recently proposed dictionary-based stochastic expectation maximization approach (developed for a medium-resolution SAR) to very high-resolution (VHR) satellite imagery, and enhanced by introduction of a novel procedure for estimating the number of mixture components, that permits to reduce appreciably its computational complexity. The specific interest is the estimation of heterogeneous statistics, and the developed method is validated in the case of the VHR SAR imagery, acquired by the last-generation satellite SAR systems, TerraSAR-X and COSMO-SkyMed. This VHR imagery allows the appreciation of various ground materials resulting in highly mixed distributions, thus posing a difficult estimation problem that has not been addressed so far. We also conduct an experimental study of the extended dictionary of state-of-the-art SAR-specific pdf models and consider the dictionary refinements.

Change-detection methods represent powerful tools for monitoring the evolution of the Earth's surface. In order to optimize the accuracy of the change maps, a multiscale approach can be adopted that jointly exploits observations at coarser and finer scales. In this letter, a multiscale contextual unsupervised change-detection method is proposed for optical images. It is based on discrete wavelet transforms and Markov random fields. Wavelets are applied to the difference image to extract multiscale features, and Markovian data fusion is used to integrate both these features and the spatial context in the change-detection process. Expectation-maximization and Besag's algorithms are used to estimate the model parameters. The selection of the optimal wavelet-transform operator within a predefined dictionary is automated by a minimum-energy criterion. Experiments on real optical images point out the effectiveness of this method as compared with state-of-the-art techniques.

Data-assimilation methods play a crucial role for exploiting remote sensing in dynamic physical models for the prediction of hydrological-process evolution. Here, a novel method is proposed to assimilate land-surface temperature estimates, derived by applying support-vector regression to infrared satellite data, into a variational technique for mass and energy exchange estimation at the soil surface. Recent techniques to fully automate support vector regression and to estimate the pixelwise statistics of the regression error are incorporated in the proposed method.

Change detection represents an important tool in environmental monitoring and disaster management. Here, a novel unsupervised change-detection method is proposed for very high-resolution SAR images, by integrating wavelet multiscale feature extraction, Markov random fields for contextual modeling, and generalized Gaussian models. Experiments with COSMO-SkyMed data remark the effectiveness of the method as compared with previous methods.

In the framework of remote-sensing image classification support vector machines (SVMs) have recently been receiving a very strong attention, thanks to their accurate results in many applications and good analytical properties. However, SVM classifiers are intrinsically noncontextual, which represents a severe limitation in image classification. In this paper, a novel method is proposed to integrate support vector classification with Markov random field models for the spatial context, and is validated with multichannel SAR and multispectral high-resolution images. The integration relies on an analytical reformulation of the Markovian minimum-energy rule in terms of a suitable SVM-like kernel expansion. Parameter-optimization and hierarchical clustering algorithms are also integrated in the method to automatically tune its input parameters and to minimize the execution time with large images and training sets, respectively.

A novel automatic method is developed for the detection of features in planetary images. Although many automatic feature extraction methods have been proposed for for remote sensing images of the Earth, these methods are typically unfeasible for planetary data that generally present low contrast and uneven illumination characteristics. Here, a novel technique for crater detection, based on a marked point process, is proposed. The main idea behind marked point processes is to model objects within a stochastic framework: They provide a powerful and methodologically rigorous framework to efficiently map and detect objects and structures in an image with an excellent robustness to noise. These methods are new and promising: They represent the last frontier of the stochastic image modeling. They have been used in different areas of the terrestrial remote sensing, but have not been applied to planetary image analysis yet. The proposed method for crater detection has many other areas applications. One such application area is image registration by matching the extracted features.

The problem of the classification of snow-covered areas from multispectral images is addressed in this paper. The key idea of the proposed technique is to integrate a decision tree classifier (DTC) and a Bayesian unsupervised thresholding algorithm, aiming at a complete automation of the classification process. Given a classification problem, the DTC approach decomposes the problem in a suitable tree-structured collection of binary sub-problems, for which simple (e.g., threshold-based) decision rules can be defined. The proposed strategy, by adopting the tree classification, discriminates several snow-covered and non-snow-covered classes, by decomposing the related multi-class problem into a set of binary thresholding sub-problems involving the multispectral channels and the resulting normalized difference vegetation index and normalized difference snow index. Focusing on a critical node in the tree, a Bayesian approach is used to expresses the threshold-selection problem as the minimization of a functional related to the probability of classification error. Experiments are reported on MODIS data.

Very high spatial resolution (HR) data provide plenty of detailed information about the ground on a regular basis for applications such as urban planning, precision farming, or damage assessment after environmental disasters. The complex nature of HR observations, especially when acquired over urban/artificial environments, makes the accurate discrimination of distinct thematic classes a difficult task. In the present paper, a novel technique is proposed for supervised classification of multispectral HR images, based on the key-idea to fuse through a Markov random field (MRF) the information conveyed by user-defined thematic classes, subclasses related to the spectral responses of different ground materials, and spatial edges. The method is validated by experiments on IKONOS images.

This paper presents a new approach for unsupervised change detection in pairs of Synthetic Aperture Radar (SAR) images. As changes to detect can have various sizes and intensities which are a priori unknown in most applications, we propose a multiscale approach without considering any a priori information. Using multiscale series of a cumulant-based Kullback-Leibler divergence (CKLD) measure computed between two dates, changes are characterized as areas where the CKLD values vary a lot when the scale varies. In a probabilistic a-contrario framework, a measure of meaningfulness of such an evolution through scale is derived, leading to a criterion free of parameter. Results are presented using a pair of SAR images acquired before and after the volcanic eruption of the Nyiragongo in January 2002 (Congo), showing the robustness of the method with respect to the number of false alarms.

Change-detection represents a powerful tool for monitoring the evolution of the Earth's surface by multitemporal remote-sensing imagery. Here, a multiscale approach is proposed, in which observations at coarser and finer scales are jointly exploited, and a multiscale contextual unsupervised change-detection method is developed for optical images. Discrete wavelet transforms are applied to extract multiscale features that discriminate changed and unchanged areas and Markovian data fusion is used to integrate both these features and the spatial contextual information in the change-detection process. Unsupervised statistical learning methods (expectation-maximization and Besag's algorithms) are used to estimate the model parameters. Experiments on burnt-forest area detection in multitemporal Landsat TM images are presented.

Land and sea surface temperatures are important input parameters for many hydrological and meteorological models. Satellite infrared remote sensing is an effective tool for mapping these variables on regional and global scales. A supervised approach, based on support vector machines (SVMs), has recently been developed to estimate surface temperature from satellite radiometry. However, in order to integrate temperature estimates into hydrological or meteorological data-assimilation schemes (e.g., in flood-prevention applications), a further critical input is often required in the form of pixelwise error statistics. This information is important because it quantifies inaccuracies in the temperature estimate computed for each pixel. This letter proposes two novel methods to model the statistics of the SVM regression error on a pixelwise basis. Both approaches take into account the nonstationary behavior of the error itself. This problem has been only recently explored in the SVM literature through the use of Bayesian reformulations of SVM regressions. The methods proposed in this letter extend this approach by integrating it with either maximum-likelihood or confidence-interval supervised estimators. In both cases, the goal is improved modeling of the error contribution due to intrinsic random variability in the data (e.g., noise). The methods are experimentally validated on Advanced Very High Resolution Radiometer (AVHRR) and Meteosat Second Generation-Spinning Enhanced Visible and Infrared Imager (MSG-SEVIRI) images.

Land surface temperature (LST) and sea surface temperature (SST) are important quantities for many environmental models. Remote sensing is a source of information for their estimation on both regional and global scales. Many algorithms have been devised to estimate LST and SST from satellite data, most of which requireaprioriinformation about the surface and the atmosphere. A recently proposed approach involves the use of support vector machines (SVMs). Based on satellite data and correspondinginsitumeasurements, they generate an approximation of the relation between them, which can subsequently be used to estimate unknown surface temperatures from additional satellite data. Such a strategy requires the user to set several internal parameters. In this paper, a method is proposed for automatically setting these parameters to quasi-optimal values in the sense of minimum estimation errors. This is achieved by minimizing a functional correlated to regression errors (i.e., the ldquospan-boundrdquo upper bound on the leave-one-out (LOO) error) which can be computed by using only the training set, without need for a further validation set. In order to minimize this functional, Powell's algorithm is adopted, since it is applicable also to nondifferentiable functions. Experimental results yielded by the proposed method are similar in accuracy to those achieved by cross-validation and by a grid search for the parameter configuration which yields the best test-set accuracy. However, the proposed method gives a dramatic reduction in the computational time required, particularly when many training samples are available.

In applications related to environmental monitoring and disaster management, multichannel synthetic aperture radar (SAR) data present a great potential, owing both to their insensitivity to atmospheric and Sun-illumination conditions and to the improved discrimination capability they may provide as compared with single-channel SAR. However, exploiting this potential requires accurate and automatic techniques to generate change maps from (multichannel) SAR images acquired over the same geographic region in different polarizations or at different frequencies at different times. In this paper, a contextual unsupervised change-detection technique (based on a data-fusion approach) is proposed for two-date multichannel SAR images. Each SAR channel is modeled as a distinct information source, and a Markovian approach to data fusion is adopted. A Markov random field model is introduced that combines together the information conveyed by each SAR channel and the spatial contextual information concerning the correlation among neighboring pixels and formulated by using ldquoenergy functions.rdquo In order to address the task of the estimation of the model parameters, the expectation-maximization algorithm is combined with the recently proposed ldquomethod of log-cumulants.rdquo The proposed technique was experimentally validated with semisimulated multipolarization and multifrequency data and with real SIR-C/XSAR images.

Multispectral and synthetic aperture radar (SAR) images are known to exhibit complementary properties: unlike optical sensors, SAR provides information about the soil roughness and moisture, and acquires useful data despite clouds and Sun-illumination conditions. However, the analysis of the resulting images turns out to be more difficult, as compared to the use of optical imagery, due to the noise-like speckle phenomenon. In order to exploit this complementarity for classification purposes, a criticality relies in the definition of accurate joint optical-SAR statistical models, due to the different physical natures of these two data typologies and to the corresponding differences in the related parametric models. In this paper, a region-based semiparametric classification technique is proposed for multisensor optical-SAR images. The method combines the tree-structured Markov random field approach to segmentation with the dependence tree approach to probability density estimation and with case-specific bivariate models for the distributions of optical and SAR data. A Bayesian decision rule is formulated at the segment level in order to incorporate spatial-contextual information and to gain robustness against noise.

Land surface temperature (LST) and sea surface temperature (SST) are important quantities for many hydrological and meteorological models and satellite infrared remote sensing represents a feasible way to map them on global and regional scales. However, in order to integrate temperature estimates into data-assimilation schemes (e.g., in applications such as flood prevention), a further critical input is often represented by the statistics of the temperature regression error. A supervised approach, based on support vector machine (SVM), has recently been developed to estimate LST and SST from satellite radiometry. In this paper, two novel methods are proposed to model the statistics of the SVM regression error occurring on each image sample. This problem has been only recently explored in the SVM literature by developing Bayesian reformulations of SVM regression. The methods proposed in this paper extend this approach by integrating it with either maximum-likelihood or confidence-interval supervised estimators in order to improve the accuracy in modelling the error contribution due to intrinsic data variability (e.g., noise).

Very high spatial resolution (HR) data provide plenty of detailed information about the ground on a regular basis for applications such as urban planning, precision farming, or damage assessment after environmental disasters. However, the complex nature of the HR observations, especially when acquired over urban/artificial environments, makes the accurate discrimination of distinct thematic classes a difficult task. In the present paper, a novel technique is proposed for supervised classification of multispectral HR images, that is based on the key-idea to combine the Markov random field (MRF) approach to data fusion with a graph-based approach to image classification in a multiscale strategy. A multiscale segmentation method is adopted in order to jointly exploit the capability to detect large structures in coarse-scale observations and to refine the identification of spatial details in fine-scale observations. Then, a novel MRF model is proposed that fuses the information conveyed by the segmentation maps at all scales and by the spatial neighborhood of each pixel. The method is validated by experiments on spaceborne and airborne HR images.

A new preprocessing technique is presented in this paper to automatically highlight changes in multitemporal strongly heterogeneous remotely sensed images. The proposed technique is devoted to the case where the two acquisitions, before and after a given event, are significantly different, due, for instance, to different sensors, acquisition modalities, or climatic conditions. In a previous study, it was proven that the local statistics of the images acquired at the two dates could be used to extract a relevant change indicator. Nevertheless, this measure is valid when the two observations have been derived from similar acquisitions. When the acquisition modalities differ, local statistics tend to be too different from one image to the other one to be relevant in highlighting the ground evolution without mixing with the changes at ground. The technique proposed in this paper to overcome this limitation is based on the assumption that some dependence indeed exists between the two images in unchanged areas. This dependence is modeled by quantile regression applied according to the copula theory and used to perform an estimation of the local statistics that would have been observed if the acquisition conditions of the first image had been similar to the ones of the second image. The method yields an estimate of the local statistics of the first image through the point of view of the second one. Then, usual Kullback-Leibler-based comparisons of those statistics are applied to define a change measure, which may be analyzed (e.g., by thresholding) in order to detect changes. Experimental results are shown to validate the proposed method by using a pair of Synthetic Aperture Radar (SAR) images onboard European Remote Sensing (ERS) Satellite images and a pair of optical-SAR images (from the High Resolution Visible (HRV) sensor onboard Satellite Pour l'Observation de la Terre (SPOT) satellite and from ERS-SAR) acquired before and after a flood.

Most existing multitemporal change detection methods use the spectral information alone. However, the inclusion of spatial and temporal information in change detection could improve the accuracy of change detection. This study proposes a new method which includes the multivariate texture in change detection by the direct multitemporal classification. The multivariate texture was extracted from two multispectral images, by the Pseudo Cross Multivariate Variogram (PCMV), which is an extension of the traditional Pseudo Cross Variogram (PCV). The experiments showed that the inclusion of multivariate texture could significantly improve the overall accuracy of change detection, compared to that of using the spectral information alone.

A new preprocessing technique is presented in this paper to perform automatic change detection in multitemporal multimodal remotely sensed images, mainly synthetic aperture radar (SAR) ones. This technique is dedicated to the case where the two acquisitions, before and after an major disaster, are different for some reason (different sensor, modality of acquisition or climatic conditions). A measure, based on the local statistics of the images between the two dates, has proved to be a relevant change indicator. Nevertheless, the measure is valid when the two observations have been acquired with a similar point of view only. When the modalities of acquisition differ, local statistics tend to be too different, from one image to the other, to be relevant to the ground evolution without mixing to the normal changes. The technique, that overcomes this constraint, is based on the assumption that some dependence exists indeed between the two images. This dependence is modelled by the copula theory and used to perform an estimation of the local statistics that would have been observed if the modality of the first image had been similar to the other. It yields an estimation of local statistics of the first image, through the point of view of the latter. Then, usual comparison of those statistics may be applied to perform change detection. Some results are shown on a pair of ERS images and pairs of SPOT/ERS acquired before and after a flood.

In the contexts of environmental monitoring and disaster management, multichannel synthetic aperture radar (SAR) data present a good potential, thanks both to their insensitivity to atmospheric and Sun-illumination conditions, and to the improved discrimination capability they may provide as compared to single-channel SAR. In this paper an unsupervised contextual change-detection method is proposed for two-date multichannel SAR images, by adopting a data-fusion approach. Each SAR channel is modelled as a distinct information source and Markovian data fusion is used by introducing a suitable Markov random field model. The task of the estimation of the model parameters is addressed by combining the expectation- maximization algorithm with the recently proposed "method of log-cumulants." The proposed technique is experimentally validated on SIR-C/XSAR data.

Multichannel synthetic aperture radar (SAR) data present a good potential for environmental monitoring and disaster management, owing both to their insensitivity to atmospheric and sun-illumination conditions, and to the improved discrimination capability they may provide as compared to single-channel SAR. However, this requires accurate and possibly automatic techniques to generate change maps from multichannel SAR images acquired from the same geographic area at different times. In this letter, an automatic unsupervised contextual change-detection method is proposed for two-date multichannel SAR images, by integrating a SAR-specific extension of the Fisher transform with a variant of the expectation-maximization algorithm and with Markov random fields. The method is validated by experiments on SIR-C/XSAR data

This paper proposes a procedure to extract spectral channels of variable bandwidths and spectral positions from the hyperspectral image in such a way as to optimize the accuracy for a specific classification problem. In particular, each spectral channel ("s-band") is obtained by averaging a group of contiguous channels of the hyperspectral image ("h-bands"). Therefore, if one wants to define m s-bands, the problem can be formulated as the optimization of the related m starting and m ending h-bands. Toward this end, we propose to adopt, as an optimization criterion, an interclass distance computed on a training set and to generate a sequence of possible solutions by one of three possible search strategies. As the proposed formalization of the problem makes it analogous to a feature-selection problem, the proposed three strategies have been derived by modifying three feature-selection strategies, namely: 1) the "sequential forward selection", 2) the "steepest ascent," and 3) the "fast constrained search". Experimental results on a well-known hyperspectral data set confirm the effectiveness of the approach, which yields better results than other widely used methods. The importance of this kind of procedure lies in feature reduction for hyperspectral image classification or in the case-based design of the spectral bands of a programmable sensor. It represents a special case of feature extraction that is expected to be more powerful than feature selection. The kind of transformation used allows the interpretability of the new features (i.e., the spectral bands) to be saved

In this paper, a novel Markov Random Field (MRF) model is proposed for the classification of high spatial resolution data. The key idea is that each thematic class can be described by a set of clusters, which represent spectral subclasses related to material responses. Accordingly, the proposed MRF model describes the contextual information by using the relationship between the (user-defined) thematic classes and the clusters extracted by a clustering procedure. This approach permits one to avoid penalizing configurations of cluster labels of neighboring pixels that correspond to the same thematic label. The proposed model has been validated on a real data set acquired by an IKONOS sensor.

In this paper we focus on the algorithm development allowing an improvement of land surface temperature (LST) estimates obtained from passive remote sensing images. An innovative regression algorithm is presented and results are compared with the main classical algorithms for LST estimation: the Split Window techniques. Specifically, a functional approximation scheme based on Support Vector Machines (SVMs) will be adopted and applied to AVHRR data. Particular attention will be devoted to the optimization of the SVM estimator, focusing on the selection of the employed kernel functions (linear, RBF gaussian, and hyperbolic tangent "tanh"). Results, suggesting the effectiveness of the proposed SVM-based approach, are presented and discussed.

The availability of synthetic aperture radar (SAR) data offers great potential for environmental monitoring due to the insensitiveness of SAR imagery to atmospheric and sunlight-illumination conditions. In addition, the short revisit time provided by future SAR-based missions will allow a huge amount of multitemporal SAR data to become systematically available for monitoring applications. In this paper, the problem of detecting the changes that occurred on the ground by analyzing SAR imagery is addressed by a completely unsupervised approach, i.e., by developing an automatic thresholding technique. The image-ratioing approach to SAR change detection is adopted, and the Kittler and Illingworth minimum-error thresholding algorithm is generalized to take into account the non-Gaussian distribution of the amplitude values of SAR images. In particular, a SAR-specific parametric modeling approach for the ratio image is proposed and integrated into the thresholding process. Experimental results, which confirm the accuracy of the method for real X-band SAR and spaceborne imaging radar C-band images, are presented

In remotely sensed data analysis, a crucial problem is represented by the need to develop accurate models for the statistics of the pixel intensities. This paper deals with the problem of probability density function (pdf) estimation in the context of synthetic aperture radar (SAR) amplitude data analysis. Several theoretical and heuristic models for the pdfs of SAR data have been proposed in the literature, which have been proved to be effective for different land-cover typologies, thus making the choice of a single optimal parametric pdf a hard task, especially when dealing with heterogeneous SAR data. In this paper, an innovative estimation algorithm is described, which faces such a problem by adopting a finite mixture model for the amplitude pdf, with mixture components belonging to a given dictionary of SAR-specific pdfs. The proposed method automatically integrates the procedures of selection of the optimal model for each component, of parameter estimation, and of optimization of the number of components by combining the stochastic expectation-maximization iterative methodology with the recently developed "method-of-log-cumulants" for parametric pdf estimation in the case of nonnegative random variables. Experimental results on several real SAR images are reported, showing that the proposed method accurately models the statistics of SAR amplitude data.

In the context of remote-sensing image analysis, Markov random field (MRF) models play an important role, due to their ability to integrate the use of contextual information associated with the image data in the analysis process, through the definition of suitable energy functions. However, especially when dealing with MRF models for supervised classification, the estimation of the parameters characterizing an MRF model is still an open issue, which is typically addressed by using time-expensive interactive "trial-and-error" procedures. In this paper, by focusing on a broad class of MRF models, in which the energy functions can be expressed as linear combinations of distinct contributions (e.g., representing different typologies of contextual interactions), an automatic supervised procedure for the optimization of the weight parameters of the combinations is proposed. Formulating the parameter-estimation problem as the solution of a system of linear inequalities, the problem is solved by extending to the present context the Ho-Kashyap algorithm, which is typically applied by the pattern-recognition community in computing linear discriminant functions for binary classification. The method is validated experimentally on three different (both single-date and multitemporal) datasets, which are endowed with distinct MRF models formalizing the spatial contextual information associated with a single remote-sensing image and/or the spatio-temporal contextual information associated with an image sequence

In the context of remotely sensed data analysis, an important problem is the development of accurate models for the statistics of the pixel intensities. Focusing on synthetic aperture radar (SAR) data, this modeling process turns out to be a crucial task, for instance, for classification or for denoising purposes. In this paper, an innovative parametric estimation methodology for SAR amplitude data is proposed that adopts a generalized Gaussian (GG) model for the complex SAR backscattered signal. A closed-form expression for the corresponding amplitude probability density function (PDF) is derived and a specific parameter estimation algorithm is developed in order to deal with the proposed model. Specifically, the recently proposed "method-of-log-cumulants" (MoLC) is applied, which stems from the adoption of the Mellin transform (instead of the usual Fourier transform) in the computation of characteristic functions and from the corresponding generalization of the concepts of moment and cumulant. For the developed GG-based amplitude model, the resulting MoLC estimates turn out to be numerically feasible and are also analytically proved to be consistent. The proposed parametric approach was validated by using several real ERS-1, XSAR, E-SAR, and NASA/JPL airborne SAR images, and the experimental results prove that the method models the amplitude PDF better than several previously proposed parametric models for backscattering phenomena.

The availability of synthetic aperture radar (SAR) data offers great potential for environmental monitoring due to the insensitiveness of SAR imagery to atmospheric and sun- light-illumination conditions. In addition, the short revisit time provided by future SAR-based missions will allow a huge amount of multitemporal SAR data to become systematically available for monitoring applications. In this paper, the problem of detecting the changes that occurred on the ground by analyzing SAR im- agery is addressed by a completely unsupervised approach, i.e., by developing an automatic thresholding technique. The image- ratioing approach to SAR change detection is adopted, and the Kittler and Illingworth minimum-error thresholding algorithm is generalized to take into account the non-Gaussian distribution of the amplitude values of SAR images. In particular, a SAR-specific parametric modeling approach for the ratio image is proposed and integrated into the thresholding process. Experimental results, which confirm the accuracy of the method for real X-band SAR and spaceborne imaging radar C-band images, are presented.

In the context of remote sensing image analysis, Markov random field (MRF) models are relevant image-analysis tools, thanks to their ability to integrate the use of contextual information associated to the image data in the analysis process. However, especially when dealing with supervised classification, the estimation of the internal parameters of the adopted MRF model is still an open issue, typically addressed by using time-expensive "trial-and-error" procedures. In the present paper, an automatic supervised MRF parameter optimization algorithm is proposed, that can be applied to a broad class of MRF models. The method formulates the parameter estimation problem as the solution of a set of linear inequalities, solved by extending to the present context the Ho-Kashyap algorithm, proposed to compute a linear discriminant function for binary classification. The method is validated experimentally on three different (single-date and multitemporal) data sets, endowed with distinct MRF models.

A general problem of supervised remotely sensed image classification assumes prior knowledge to be available for all the thematic classes that are present in the considered dataset. However, the ground-truth map representing that prior knowledge usually does not really describe all the land-cover typologies in the image, and the generation of a complete training set often represents a time-consuming, difficult and expensive task. This problem affects the performances of supervised classifiers, which erroneously assign each sample drawn from an unknown class to one of the known classes. In the present paper, a classification strategy is described that allows the identification of samples drawn from unknown classes through the application of a suitable Bayesian decision rule. The proposed approach is based on support vector machines (SVMs) for the estimation of probability density functions and on a recursive procedure to generate prior probability estimates for known and unknown classes. In the experiments, both a synthetic dataset and two real datasets were used.

In remotely sensed data analysis, a crucial problem is represented by the need to develop accurate models for the statistics of the pixel intensities. This paper deals with the problem of parametric probability density function (PDF) estimation in the context of Synthetic Aperture Radar (SAR) amplitude data analysis. Several theoretical and heuristic models for the PDFs of SAR data have been proposed in the literature, that have been proved to be effective for different land-cover typologies, thus making the choice of a single optimal SAR parametric PDF a hard task. In this paper, an innovative estimation algorithm is described, that faces such a problem by adopting a finite mixture model (FMM) for the amplitude PDF, with mixture components belonging to a given dictionary of SAR-specific PDFs. The method automatically integrates the procedures of selection of the optimal model for each component, of parameter estimation, and of optimization of the number of components by combining the Stochastic Expectation Maximization (SEM) iterative methodology with the recently developed "method-of-log-cumulants" (MoLC) for parametric PDF estimation. Experimental results on several real SAR images are reported, showing that the proposed method accurately models the statistics of SAR amplitude data.

Purpose of this paper is to propose a procedure to extract, from a hyperspectral image, spectral channels of variable bandwidths and spectral positions in such a way as to optimize the accuracy for a specific classification problem. In particular, each spectral channel ("s-band") is obtained by averaging a group of contiguous channels of the hyperspectral image ("h-bands"). If one wants to define n s-bands, the problem can be therefore formulated as the optimization of n starting and n ending h-bands. To this end, we propose to adopt as optimization criterion an interclass distance computed on a training set and to generate a sequence of possible solutions with one of three possible search strategies. As the proposed formalization of the problem makes it analogous to a feature selection problem, the three proposed strategies have been derived by modifying three feature selection strategies: the Sequential Forward Selection (SFS), the Steepest Ascent and the Fast Constrained Search. Experimental results with a well-known hyperspectral data set confirm the effectiveness of the approach, which allows better results than those provided by the SFS method for feature selection. A preliminary comparison suggests that the accuracy is very similar to that obtained by the DBFE feature transformation method. The interest of this kind of procedure can be for a case-based design of the spectral bands of a programmable sensor or for the reduction of the number of bands derived from a hyperspectral image. It represents a special case of feature transformation that is expected to be more powerful than feature selection. The kind of transformation used allows the interpretability of the new features (i.e. the spectral bands) to be saved.

A key-problem in dealing with multitemporal images of a given geographical area is the identification of the changes occurring between distinct acquisition dates. A complete map of the change typologies can be generated when training data are available for all observation dates, but this completely supervised context involves expensive requirements. On the other hand, a completely unsupervised context does not require any prior information but does not allow an analysis of the different typologies of change, since no class information is available at any observation date. In the present paper, a contextual multitemporal classification and change detection algorithm is proposed, which deals with remotely sensed image sequences with ground truth information available only at none reference acquisition date. The method integrates clustering information with a two-stage contextual Markov Random Field (MRF) model for the spatio-temporal correlation associated to the sequence. The algorithm is validated on a multitemporal and multispectral real data set, acquired over an agricultural and urban area, and characterized by a large amount of changes between the observation dates.

This paper addresses the problem of classifying multispectral images when the a priori knowledge about classes is not complete: the true number of classes is not known, or it is not possible to obtain ground truth data for some of the classes in the image. We propose a method to perform image classification taking into account all the classes, "known" and "unknown", based on accurate estimates of the prior probabilities and of the joint probability density functions (pdfs). To this end, we propose the application of the dependence tree approximation to mitigate the problem of few available samples. Finally, we investigate the suitability of the application of a biased cross-validation criterion for the optimization of 2-dimensional pdf estimations.

Given a temporal sequence of remote sensing images, the difficulty to collect regularly in time ground truth information makes it important to develop automatic unsupervised change detection techniques. Due to its simplicity, image differencing represents a popular approach for change detection. To separate the "change" and "no-change" classes in the difference image, a thresholding-based procedure can be applied. However, the main weakness of an unsupervised change detection approach is the absence of prior information about the scene as it resorts to the spectral information, only, which does not allow the analysis of the typologies of changes occurring between the acquisition dates. In the monitoring of a given study area, the main problem is that the ground truth collection does not usually follow the image acquisitions at the different dates. However, it is easier to have at least one image for which the ground truth is available. In this paper, we propose a partially supervised change detection scheme that is based on the exploitation of the ground truth availability for at least one temporal image. A clustering algorithm is applied to both acquired images and a thresholding-based unsupervised change detection algorithm is applied inside each cluster of the second date image in order not only to identify the presence of changes but also to distinguish between different typologies of changes.