Target Detection For SAR Image Based On Primal Sketch

Synthetic Aperture Radar (SAR) has been widely used in military reconnaissance based on its unique advantages. Usually, the military reconnaissance targets on the ground mainly include bridges, military building, vehicles and ships in SAR, so it’s meaningful and has application prospect to carry out the target detection of the SAR. Primal Sketch is a sparse representation for image which can depicts the singular information of an image such as object boundary, direction and location regarding line segments as its primitive in a form of sketch. In SAR images, there are many human objects and they are different from each other in size, for the features, In this paper, we first propose to detect targets in the SAR images using the sparse representation domain of Primal Sketch. The main contributions can be summarized as follows:At first, we study the characteristics of the SAR images and analyze the difficulty of the target detection and the theory frame of Primal Sketch and its implementation. According to the topology between the segments, we propose to use the feature regularity and so on to define the seed segments in the appropriate window, theoretical analysis and experiments show that using the regularity we can effectively capture the artificial targets in SAR images.Secondly, in the domain of the sparse representation of Primal Sketch in the SAR images, for artificial targets, we choose the appropriate scale of the window and an algorithm of the formation of the target area in the SAR images-region growing based on seed segments is proposed. Firstly, we define the attribute set of all segments, and then select the seed segments according to their attribute set. At last, region growing is made for a chosen seed segment according to the set rule. Different from the traditional algorithm of region growing, the processing unit of our algorithm is no longer the pixel but the segment. So it can preferably capture the structure information of the image. Experiments show that with our algorithm, we can preferably extract potential target area.At last, after extracting potential target area, we sort the detected potential target area by the regularity and the intensity of the area to select our interesting target area, thus finish our target detection. Experiments show that using the algorithm, we can effectively extract the artificial targets in the SAR images.