SAR Target Detection In The Complex Scene

With the widely use of the synthetic aperture radar(SAR) systems, the number of SAR images increases rapidly, which makes the automatic target detection draw extensive attentions. In the meantime, with more and more SAR systems being used in complex scenes, the study of detection algorithm that can have good detection performance in complex scenes have become an important research topic.The thesis first briefly introduces the research background and significance of SAR image target detection. Then it introduces some conventional SAR image detection algorithms and the conventional clustering algorithm, and analyzes the performances of them. Based on the existing methods, the thesis focuses on the study of target detection in complex scenes, including the targets on the ground and in the port. The main work of this thesis is summarized as follows.1. The thesis proposes a superpixel-based two-parameter constant false alarm rate(CFAR) target detection algorithm for the high-resolution SAR images. It consists of three steps: segmentation, detection and clustering. In the segmentation step, the thesis proposes a SAR image superpixel segmentation algorithm, which can segment the targets and clutter into different superpixels. In the detection step, the thesis combines the conventional two-parameter CFAR algorithm with the segmentation result. With the help of the superpixels, the clutter area for every pixel can be adaptively chosen. Also the disturbances of the adjacent targets are decreased. Therefore, the estimate of the clutter pixel distribution parameters is more accurate and the detection algorithm can achieve better detection performance. In the clustering step, the thesis proposes a superpixel-based clustering algorithm, which can distinguish the adjacent targets and has better performance in the multi-target scenes.2. The thesis studies a superpixel-based multi-feature detection algorithm. The detection algorithm consists of three steps: segmentation, feature extraction and detection. In the segmentation step, the thesis still uses the superpixel segmentation algorithm mentioned above. The segmentation can segment a target into several superpixels. In this algorithm, one target is no longer regarded as a set of pixels but some superpixels with shape and structure information. In the feature extraction step, three pixel-level detection statistics are extended to superpxiel-level, and combined as the feature of the superpixel. In the detection step, based on the prior knowledge, the support vector data description(SVDD) is used to perform the detection. This algorithm can achieve better detection performance and is more stable.3. The thesis studies a detection algorithm for the ship targets in the port. The algorithm consists of three steps: segmentation, detection and clustering. After the land and the sea are segmented, the following detection and clustering can be taken only in the sea and the port instead of the whole image, which decreases the calculation amount. In the detection stage, the cell average CFAR(CA-CFAR) is used for detection. In the clustering stage, the thesis studies a coastline angle-based clustering algorithm. The algorithm can effectively cluster the ship pixels around the port. After the clustering, the clustered clutter can be removed based on the shape information that is different for the targets and the clutter.