Research On Key Technologies For Ship Surveillance Based On HRWS SAR Imagery

Ship surveillance is one of the most important applications of synthetic aperture radar(SAR) imagery, as it can help to improve the activities of fishing control, marine traffic monitoring, anti-terrorism, and oceanic rights protection. The high resolution and wide swath(HRWS) SAR provides the advantages of high resolution and wide swath synchronously, which promotes the marine surveillance performance of SAR imagery, as well as brings additional challenges. Therefore, the research on HRWS SAR ship surveillance is an advanced issue with great theoretical and applicable significances.With the development of HRWS SAR imaging techniques, ship surveillance on HRWS SAR imagery confronts novel challenges. Therefore, this dissertation mainly focuses on serveral of the key techniques to improve the ship surveillance performance, including the ship detection on single channel and polarimetric SAR imagery, feature extraction and classification of ships on high resolution SAR imagery. The main contributions of this dissertation comprise the following subjects:1. To improve the ship detection performances under rough environments of clutter edges and multiple interferers, the dissertation proposes a VIE-CFAR(Variable Index and Excision Constant False Alarm Rate) based ship detection algorithm on SAR imagery. Fistly, we investigate the statistical characteristics of sea clutter on SAR imagery with different radar frequencies, polarimetric modes, resolutions and sea conditions. Extensive experiments on measured SAR images validate the effectiveness of 0G model to descripe the sea clutter. Secondly, we design a VIE-CFAR detector and analysis its performances under various environments of homogeneous, clutter edges, and multiple interferers. Finaly, we propose the VIE-CFAR based ship detection algorithm under complex environment and validate it on ENVISAT, Radarsat-2 SAR images. The detection results demonstrate that the proposed method superiors to other CFAR methods for ship detection with the presence of clutter edges and multiple interferers.2. Aiming at alleviating the additional false alarms of azimuth and sidelobe ambiguities on HRWS SAR imagery, we investigate the ship detection with polarimetric SAR data, and propose a feature selection and weighted(FSW) support vector machine(SVM) detection algorithm to detect ships on polarimetric SAR imagery. At first, we analysis the conventional ship detection methods and their detection abilities on polarimetric SAR imagery. And then, we construct a multi-polarization feature vector to represent the ships, ambiguities and sea clutters on SAR imagery, and refine the feature vector to a weighted optimal one, which is imported into the SVM classifier for ship detection purpose. We conduct extensive experiments on air-borne AIRSAR and space-borne Radarsat-2 polarimetric SAR images, and the results validate that the propose method is robutness on different images and allivates the false alarms caused by azimuth and sidelobe ambiguities.3. Benefit from the high resolution property of HRWS SAR imagery, the thesis proposes algorithms to extract precise geometric features, and extract a novel superstructure scattering feature based on the ship’s structure and its scattering characteristic. Firstly, we investigate the conventional feature extraction methods of ships on SAR imagery. Secondly, to alleivate the error caused by imaging sidelobes and neighboring targets, we propose an algorithm of precise geometric feature extraction based on Radon transform and histogram analysis, and validate it on high resolution Terra SAR-X SAR images. Finally, we analysis the physical structures of typical merchant ships and the corresponding electromagnetic scattering behaviors, and propose an algorithm to extract the superstructure feature for recognition purpose. Validation on a Terra SAR-X SAR dataset consists of the Container, Oil Tanker and Bulk ships demonstrate the effectiveness of the feature.4. By introducing the sparse representation classification(SRC) theory, the thesis propose a sparse representation based SAR vehicle recognition method along with aspect angle and a ship classification method with feature space based sparse representation. Firstly, we investigate the SRC method and validate its effectiveness with MSTAR(Moving and Stationary Target Acquisition and Recognition) dataset. Secondly, based on the analysis of the correlations between target’s aspect angle and the sparse representation coefficients, we propose a sparse representation based SAR vehicle recognition method along with aspect angle. Extensive experiments on MSTAR dataset demonstrate that the proposed method performs robustly under the variations of depression angle and target configurations, as well as incomplete observation. Finally, we investigate the dictionary construction method of the SRC method on feature space, and propose a ship classification method with feature space based sparse representation. Classification results on the three kinds of typical merchant ships validate that the proposed method superiors to the template maching, K-nearst neighbor(KNN), Bayes, and SVM methods.