Matteo Soccorsi

Neural networks for oil spill detection using TerraSAR-X data

The increased amount of available Synthetic Aperture Radar (SAR) images involves a growing workload on the operators at analysis centers. In addition, even if the operators go through extensive training to learn manual oil spill detection, they can provide different and subjective responses. Hence, the upgrade and improvements of algorithms for automatic detection that can help in screening the images and prioritizing the alarms are of great benefit. In this paper we present the potentialities of TerraSAR-X (TS-X) data and Neural Network algorithms for oil spills detection. The radar on board satellite TS-X provides X-band images with a resolution of up to 1m. Such resolution can be very effective in the monitoring of coastal areas to prevent sea oil pollution. The network input is a vector containing the values of a set of features characterizing an oil spill candidate. The network output gives the probability for the candidate to be a real oil spill. Candidates with a probability less than 50% are classified as look-alikes. The overall classification performances have been evaluated on a data set of 50 TS-X images containing more than 150 examples of certified oil spills and well-known look-alikes (e.g. low wind areas, wind shadows, biogenic films). The preliminary classification results are satisfactory with an overall detection accuracy above 80%.

Huber–Markov Model for Complex SAR Image Restoration

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

Azimuth ambiguities removal for ship detection using full polarimetric X-band SAR data

Synthetic Aperture Radar (SAR) is an active sensor that provides high-resolution images of the Earth surface day and night with almost no interferences of the weather conditions. Thanks to these characteristics, during the past decade, its use in the field of maritime security and safety has increased. Nowadays one of the SAR-based applications to increase the maritime security is ship detection. Automatic SAR ship detection is not a trivial task due to the speckle and processes that may cause false alarms. Azimuth ambiguities caused by the aliasing of the Doppler phase history are often visible in low radar backscatter background causing false positive. In this work we propose a method to resolve azimuth ambiguities for ship detection purposes that takes the complete benefit of full Polarimetric SAR (PolSAR) data combined with the high resolution capability of X-band sensors.

Stochastic models of SLC HR SAR images

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

Ship Surveillance with High Resolution TerraSAR-X Satellite in African Waters

Ship detection is an important application of monitoring of environment and security or safety issues in African Waters. In order to overcome the limitations by other monitoring systems, e.g. coastal radar, surveillance with satellite Synthetic Aperture Radar (SAR) is used because of its potential to detect ships at high resolution over wide swaths and in all weather conditions and independent from sun illumination. TerraSAR-X (TS-X) is an X-band polarimetric SAR capable of imaging up to 1 m resolution in Spotlight mode. TS-X can be used for a wide variety of applications and methods of analysis including visual interpretation, mapping, digital-elevation-model creation, disaster monitoring, and oceanography. Results on the combined use of TS-X ship detection, Automatic Identification System (AIS), and satellite AIS (SatAIS) are presented. Using AIS is an effective terrestrial method for tracking vessels in real time typically up to 40 km off the coast. SatAIS is a space-based system with nearly global coverage for monitoring of AIS equipped ships. Since not all ships operate their AIS and smaller ships are not equipped with AIS, space borne SARs provide complimentary means for ship monitoring. As cases, images were acquired over the Somali Coast Area, South African Coast and Gibraltar in Stripmap mode with a resolution of 3 m at a coverage of 30 km× 50 km. The rapid tasking performance as well as the short response time of the TS-X data acquisition of the ground segment DLR-BN (Ground Station Neustrelitz, Germany), are very helpful to monitor hotspot areas such as the Gulf of Aden. For ascending orbits the delivery time of ship detection products is less than 20 min. Along with the detected ship positions, estimated wave heights and wind fields derived from large-area TS-X imagery can be used to get a detailed maritime picture of the situation.

Parametric versus non-parametric complex image analysis

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

SAR complex image analysis: A Gauss Markov and a multiple sub-aperture based target characterization

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

Automated information extraction from high resolution SAR images: TerraSAR-X interpretation applications

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

Phase characterization of polarimetric SAR images

High Resolution (HR) Synthetic Aperture Radar (SAR) Single Look Complex (SLC) observations, mainly of strong scattering scenes or objects show phase patterns. Phase patterns may occur due to the system behavior or they may be signatures of the imaged objects. Since state of the art stochastic models of SAR SLC data describe mainly the pixel information. Now studies are needed to elaborate better models for the full information content. Thus, new statistical models of HR SAR SLC are proposed, they aim at the characterization of the spatial phase feature of Polarimetric SAR (PolSAR) SLC data, i.e. they describe multi-band, complex valued textures. The definition of texture must be changed because it is not anymore characterizing the optical features but the electromagnetic properties of the illuminated targets. The content of the SAR image is a stochastic process characterized from its own structure and geometry, which differs from the real one of the illuminated scene, and is dominated from strong scatterers. Nevertheless we are going to accept the classical texture definition, inherited from computer vision, in homogeneous areas and, furthermore, we are going to extend it for a characterization of isolated and structured objects The proposed models are in the class of simultaneous Auto-Regressive (sAR) defined on a generalized set of cliques in the pixel vicinity. Models may have different orders, thus capturing different degrees of the data complexity. To cope with the problem of estimation and model order selection Bayesian inference is used. The results are presented on PolSAR data.

Space-Variant Model Fitting and Selection for Image Information Extraction

With the growing importance of model‐based signal analysis methods, the dependence of their performance on the choice of the models needs to be addressed. Bayesian theory incorporates model selection in a natural and direct way: we apply it to the space‐variant choice of the best model in a given reference class in the framework of parameter estimation from complex data. In particular, we introduce an algorithm for image information extraction that is based on a two‐level model, it estimates local texture Gauss‐Markov Random Field (GMRF) parameters and local GMRF model order for incomplete data. Model selection is based on an approximate numerical computation of the evidence integral. Results are presented on Synthetic Aperture Radar (SAR) images.