With the increase of high spatial and temporal resolution SAR data availability, novel applications and information extraction techniques become possible. Among these, the extraction of micro-motion information has the potential to unlock a range of applications, such as infrastructure monitoring, maritime surveillance and natural disaster damage assessment. However, sensors, acquisition modes and products have not been designed with in mind the optimization of micro-motion extraction and its applications, therefore, careful considerations need to take place when selecting the most suitable data and designing processing algorithms. In this paper practical and processing considerations when dealing micro-motion extraction from high-resolution SAR sensors are discussed and supported with experimental results obtained from Capella, Umbra and TerraSAR-X data.

Phase unwrapping is a key problem in several technical fields, among which SAR interferometry (InSAR). In this work, a new computationally efficient approach to spatio-temporal three-dimensional (3D) phase unwrapping is proposed. The method makes it feasible to process very large datasets fully exploiting the information from the spatial and temporal structures of the data. The global 3D phase unwrapping problem is separated into smaller subproblems, each of them global spatially or temporally, which are solved iteratively till convergence to a suboptimal but still very good solution. Moreover, to further improve computational efficiency, we propose an iterative least-squares strategy for phase unwrapping that exploits the integer character of the phase ambiguities to be recovered and the repeated structures in the subproblems to be solved. As a third element, which can additionally improve computational performance and result robustness, a pyramidal solution strategy can be considered, consisting of successive phase unwrapping of subsets of points, typically with increasing noise, each constrained by the previous results. The tests performed on simulated and real satellite InSAR data confirm the validity of the proposed approach.

This work consolidates the existing results in the field of information extraction from spaceborne SAR imagery based on Inverse Synthetic Aperture Radar (ISAR) techniques, as well as enhances the understanding of the phenomenology, the models, the processing algorithms, the applications, and the overall value in security applications. The focus is on ISAR based advanced processing methods to explore the potentialities of very high resolution (VHR) SAR data in a range of security related application domains, including maritime, inland water, and land scenarios.

Synthetic aperture radar (SAR) interferometry (InSAR) is a powerful technology to monitor from satellite very large areas and detect motions of the ground surface (typically due to subsidence, landslides, earthquakes, and volcanic phenomena) with millimetric precision and sub-metric spatial detail (making it possible to distinguish different parts of buildings or infrastructures). A key step of this technology is the identification of points (or clusters of points) providing a coherent backscattering over time. These points typically correspond to man-made structures, rocks, or bare soil, and can be called persistent scatterers (PSs) regardless of whether the dominant diffusion mechanism is point-like or distributed. In this work, we propose a novel algorithm, which we will call point coherence estimation (PCE), to evaluate in a clean and simple way (without the need for articulated procedures and critical assumptions or approximations) the interferometric coherence of each single point in an interferometric image series. The method exploits the well-accepted assumptions that large phase artefacts such as atmospheric delays or orbital effects are almost identical between points within tens or hundreds of meters, whereas phase noise (e.g., thermal noise and time, spectral, geometric decorrelation noises) is relatively small w.r.t. a phase cycle and has statistically independent realizations in different points (possibly excluding adjacent pixels if the images are oversampled). Based only on these assumptions, it is possible to write an overdetermined system of linear equations to evaluate the coherence of each single point reliably and consistently from the coherences of pairs of points, which can be calculated directly. The tests performed on simulated and real datasets confirm the validity and great potential of the method.

This paper presents a novel processing chain based on Synthetic Aperture Radar (SAR) and Inverse SAR (ISAR) techniques to refocus SAR images of moving maritime targets and estimate their motion parameters. The proposed processing chain was developed and extensively evaluated The algorithm was tested on a large dataset of COSMO-SkyMed (CSK) and Cosmo Second Generation (CGS) dataset including 300 vessels, and spanning different maritime scenarios, in order to account for many contingencies such as the sea states, the ship movements, and the mutual geometry between the SAR orbit and the course of ship. To assess the refocusing capability, quantitative contrast measurements of the refocused images were conducted. The accuracy of vessel speed estimation was evaluated by comparing the results with Automatic Identification System (AIS) data as a reference. The key contribution of this work lies in demonstrating the effectiveness of ISAR processing applied to satellite SAR images for near real-time Maritime Awareness applications. This was achieved through the development of a robust and fully automatic processing chain, as well as an extensive experimentation and validation process.

Ship detection using synthetic aperture radar images is a key technology in maritime surveillance applications. In addition to the position of the vessel, the characterization of the target (length, width and orientation) is often a requirement. In this paper, we present a deep learning architecture for object detection we developed by modifying the popular YOLOv3 architecture to apply to vessel detection and parameter estimation from SAR images. The proposed architecture was trained and tested on a large dataset of SAR images defined in this work. It contains images covering a wide range of spatial resolutions (pixel spacing ranging from 1.5m to 50m) and labelled with oriented bounding boxes to associate to each vessel not only its position but also size and orientation. The obtained results are very promising and confirm the validity of the approach.

The current remote sensing (RS) open data policy for multispectral (MS) missions such as Sentinel-2 and Landsat-8, together with the availability of free cloud distributed processing platforms such as Google Earth Engine, makes it possible the quick generation of burned area (BA) products even for nonexperts in the field. Indeed, fires and BAs can be detected using burn severity indices, which are usually obtained by simple band algebra operations. However, simple approaches can aid BA estimation only if typical error patterns are known and accounted for, especially when working at large (e.g., continental) scales. This article proposes an automatic BA detection system based on burn severity index thresholding, which integrates dedicated false and missed alarm mitigation strategies to improve the detection accuracy. The system is tested on Sentinel-2 and Landsat-8 data over ten different locations in Europe and spanning year 2018. Three known burn severity indices plus a custom one defined to improve the performance in the considered study area are under study. Results show that burned index thresholding is possible within accuracy bounds slightly larger than the state of the art, which is acceptable by considering the proposed simplified processing framework.

The increasing availability of huge amounts of satellite data, together with the increasing computation power available at relatively low cost, is requiring and at the same time allowing the development of new algorithms to automatically extract information from the data. In this work, we propose a new method for automatic detection, from series of satellite images, of possible anomalies relative to crop parcels declared by farmers in the framework of EU's Common Agricultural Policy (CAP). Differently from other recently explored methods, our technique is not based on a crop classification approach. On the contrary, we approached the problem as an anomaly detection problem, and our method bases only on the quite realistic and general assumption that declarations are mostly correct, with a moderate number of outliers. Therefore, our technique is robust to variations in weather conditions, terrain morphology and agriculture practices. In order to detect the anomalies, the method computes the “distances” between the different parcels with a given declared crop. In particular, the time series of the features extracted from the satellite data on the different parcels are compared. Their distance is defined according to the Dynamic Time Warping (DTW) method, robust to temporal variations. The tests performed were very good, and the technique has been already operationally used with satisfactory results. In particular, the automatic anomaly detection approach has made it possible to verify all the farmers' declarations in a large area (a significant portion of Italy), and will make it possible to process even larger areas, such as the whole Italy or the whole Europe.

Ship detection SAR images is a key technology in maritime surveillance applications. We show here the main issues arising in the processing SAR data for ship detection and how techniques, based on Constant False Alarm Rate (CFAR) algorithms, have evolved to face them and to achieve operational performances. This evolution followed the advancements in the satellite SAR system technology that have increased acquisition capability in terms of always greater number of images and modes, better spatial resolutions and wider swaths. We describe an operational CFAR ship detection algorithm having some novel features with respect to CFAR algorithms available in literature to improve quality, robustness and processing time.

Grassland use intensity has an impact on their ecological value as habitats. The precocity and frequency of mowing events are major factors of grassland use intensity. Grassland mowing detection through remote sensing can thereby be a great asset for large scale habitat monitoring. A grassland mowing product, based on Sentinel-1 and Sentinel-2 time series, was developed recently in the frame of ESA's Sentinels for Common Agricultural Policy (Sen4CAP) project. The aim of this study is to assess the performances of this Sen4CAP mowing algorithm on managed grasslands in Belgium. Based on a large reference data set, collected through field observations in 2019, this study shows that the product detects 79% mowing events in managed grasslands and that its confidence level estimation is strongly correlated to the detection precision. Overall, the Sen4CAP grassland mowing product represents a great potential for grassland use intensity assessment in the context of large scale monitoring of biodiversity habitat.

Oil spills, caused by accidents or by ships cleaning their tanks, represent big threats for maritime and coastal ecosystems health. A very effective detection of oil spills can be performed using satellite synthetic aperture radar (SAR) systems, operating regardless of cloud coverage and sunlight and capable of discriminating oil from regular sea surface. However, discriminating between real oil spills and lookalikes (such as natural oils and seepages, often occurring in upwelling sea areas), although well performed by expert SAR image interpreters, poses a great challenge for automatic processes. In addition, a visual check performed by human operators on a great number of images would be too expensive. Therefore, many solutions for automatic detection have been tried in the last few years, using probabilistic models and, more recently, machine learning. This work presents an innovative solution based on image-to-image translation using convolutional neural networks (CNNs) trained with an adversarial loss function. The proposed approach has been tested, with very promising results, using Radarsat-2 and Sentinel-1 SAR data over the Mediterranean Sea and some areas of the Atlantic Ocean and the North Sea.

Remote sensing present some new challenges for deep learning, because (also to compensate the scarce detail level) multimodal, multisource and multitemporal data should be jointly exploited. For example, time series of optical multispectral/hyperspectral or synthetic aperture radar (SAR) data probe different properties of the observed scene, based on their different wavelength, acquisition geometry, etc., and with possible data gaps. To address this task, we propose a new deep learning architecture that exploits a sequence of deep convolutional neural networks (CNN) and a recurrent neural network (RNN). In the proposed architecture, all the data (with their spectral, spatial and temporal information) are used jointly and optimally in the sense that no imputation is enforced, but the internal weights providing the best classification results are estimated from the data themselves (hence the proposed name ODIN - Optimal Data Imputation Network). We have tested the proposed architecture, using Sentinel SAR and multispectral image series, on land cover and crop classification, an important remote sensing application. The obtained results are very promising, with an error rate below 1%, and show good spatial consistency without loss of spatial resolution.

Image coregistration aims at stacking two or multiple images in a way such that, for each image, the same pixel corresponds to the same point of the target scene (possibly with sub-pixel accuracy). We can distinguish two families of image coregistration problems, basically depending on if the images to be coregistered are taken by sensors of the same or different type (e.g., sensing different wavelenghts), and with similar or different illumination and acquisition geometries (e.g. different sun illumination conditions and/or different acquisition incidence angles). Whilst the first type of image coregistration is well established, multimodal coregistration is not yet well founded and due to difficulty of finding correspondences between the images (tie points) in a robust way, and the avable approaches often recur to manual assistance. The multimodal image coregistration technique proposed in this work overcomes the problems due to differences in radiometries and in geometries by exploiting two main concepts: complementary geometry information between the images to be coregistered, and mutual information (or entropy) as similarity metric. The method focuses on coregistration of very high resolution synthetic aperture radar (SAR) and optical images, but the approach is of general validity. The tests performed on real very high resolution optical and SAR data confirm the validity of the method.

Due to the finite pulse repetition frequency and non-ideal antenna pattern, the presence of “ghosts” on SAR images of maritime scenes is frequently observed. This phenomenon can lead to an increase in false alarm rate in ship-detection applications. In this paper we propose to use the recently developed “asymmetric mapping and selective filtering” (AM&SF) method for the filtering of azimuth ambiguities on stripmap SAR images as a preliminary step of an adaptive-threshold cell-averaging constant-false-alarm-rate ship-detection algorithm. We show that use of this preliminary filtering step allows us to significantly improve the performance of the ship detection by reducing the false alarm rate, without reducing the detection rate. The proposed framework is positively applied to a couple of Cosmo/SkyMed SAR images.

Image segmentation is the problem of finding the homogeneous regions (segments) in an image. Applications of image segmentation range from filtering of noisy images to problems of feature extraction and recognition. In this work we present a novel approach for image segmentation problems. The proposed technique is based on the idea of splitting the original image segmentation problem in two subproblems with lower computational complexity. First, a preliminary estimate of the segmented image gradient is found by solving a number of one-dimensional segmentation problems. In a second step, the results are merged together by enforcing that the obtained vector field is irrotational. At the cost of obtaining a "sub-optimal" solution, the computational advantage coming from the proposed decomposition can allow the implementation of sophisticated strategies that would be practically impossible to implement in a unique step. The results obtained on real and simulated image confirm the validity of the proposed approach.