Study On Ship Detection Within Harbor Areas In SAR Images

Ship detection in synthetic aperture radar(SAR) images, which is of important significance for military information collection, ocean surveillance and fishing monitoring, has become a hot research topic in the field of ocean remote sensing. Since ships are of high density and frequent movement in harbor areas, especially in some important harbors, with a demand of valuable observation, the study on the ship detection within harbor areas in SAR images is of great realistic significance. Aiming at accurate extraction of ship targets from SAR images, the thesis focuses on the technologies of sea-land segmentation, ship target detection and discrimination in SAR images.In essence, ship detection can be seen as a stepwise data screening procedure. For SAR images, it includes three steps, sea-land segmentation, target detection and false alarm discrimination, where sea-land segmentation is to remove land areas, target detection is to extract the ROIs, possible to be ship targets, from the ocean, and false alarm discrimination is to eliminate false alarms from the detection result and output the real ship targets finally. According to this train of thought, our work in the thesis is explored as follows:The background of the harbor area is comparatively complicate in SAR images. Since docks and ships are both strong scattering targets and have similar gray values, a ship appears to be integrated with the dock when it is at anchor in the neighborhood. As a result, traditional detection methods can hardly separate the ship from the dock. To solve this problem, the thesis proposes a new ship detection method based on sea-land segmentation. Using the optical image and the corresponding harbor mask as the prior knowledge, this method casts the optical mask upon the SAR image through an auto-registration algorithm between the optical and SAR images and achieves the separation of the ship and the dock. Then the global CFAR algorithm is adopted to extract ship targets rapidly in the ocean area with a homogeneous background.In the discrimination procedure, discrimination based on features is most widely used. According to the difference between ship targets and sea clutters, the thesis introduces a new discrimination feature based on the change detection technique, named as the target pixel aggregative measure(TPAM). TPAM can assess the aggregation of pixels of a strong scattering target in the target region, to discriminate targets and clutters.Besides, geometric feature is also an important feature for ship discrimination. Nevertheless because of coherent imaging, outspread shadows and crossed sidelobes often appear around ship targets in SAR images, which bring difficulties to extracting the geometric feature of ship targets. According to this problem, the thesis proposes a method for geometric feature extraction of ship target which is based on an ellipse fitting algorithm by using the similarity between a ship contour and an ellipse. The experiments with real SAR data and verify the validity of the proposed method.