Target Detection For SAR Image Based On Sketch Sparse Representation And Low-Rank Decomposition

In the process of Synthetic Aperture Radar(SAR) image target detection,we extract interest areas within potential targets first,then positioning artificial objects in these potential target areas to realize the separation of target and background and compelete the process of target detection. As a result of SAR in aerospace, ground battlefield surveillance, reconnaissance, weapon guidance, and other fields has important application value, the target detection based on SAR image has become one of the core technology of the application of SAR in civilian and military and is of great significance on the subsequent target identification and classification. SAR sketch is a kind of sparse description of image structure information. Edge-line model based on ratio and correlation operator can describe singular information in SAR images more better. Low-rank decomposition based on the constraint of low rank background and sparse foreground in observation matrix, the target and noise exist in said prospects of sparse matrix after processing.In this paper, a sketch sparse representation based candidate target area extraction method and a target area and low-rank decomposition based SAR image artificial target positioning mothod has been proposed, which integrates SAR sketch model and robust principal component analysis. The main innovation can be summarized as follows:(1) In view of problems of artificial target positioning inaccurate, based on the structure and decomposition of the observation matrix to the artificial target, a target area and low-rank decomposition based SAR image target detection algorithm has been proposed. Firstly, we research and analysis segments regularity in sketch graph based candidate target area extraction algorithm that is proposed by Liu Fang and Song Jianmei. But this method adopt sketch graph which extract through the Primal Sketch. The edge detection operator in this model is designed in view of the optical image additive noise, which do not apply to the multiplicative noise model in SAR. Therefore, we extract candidate target areas on the basis of sketch map through the SAR sketch model. Then, we eliminate false alarm target areas, design the structure of observation matrix for the target areas, decompose this matrix to low rank matrix and sparse matrix by using Rpca and realize artificial target positioning in the target area by statistical features of sparse matrix.(2) Based on the candidate target areas extraction algorithm through neat degree of the sketch graph segments and the SAR sketch model, we analysis the sparse representation features of different types of artificial targets in SAR sketch map, define the adaptive window which is used to calculate the sketch segments neat and design the rules of the regional expansion. Compared with the fixed window size and recursive growth rules in original candidate target areas extraction algorithm, the algorithm proposed in this paper based on the SAR sketch map and adaptive geometric structure window can effectively reduce the false alarm ratio in the candidate target areas, more conductive to subsequent artificial target positioning.