SAR Image Target Discrimination Algorithm

The Automatic Target Recognition (ATR) of SAR image contains3stages, they are image pre-processing stage, feature extraction and selection of target information stage and Target recognition stage. And the feature extraction and selection of target information stage is the most important one. By extracting feature of those objects who are being detected, it can filter the real aims and false aims. There are several algorithms aim at different features. This paper proposed a new algorithm by traditional algorithms, which could be easily realized. At the same time, this paper studied several different algorithms and then picked up two of them to excise. It was proved that the combination of the two algorithms was better than separated ones. And this paper referred a software for SAR-ATR.1) This paper proposed a new discrimination algorithm, RVE algorithm. With this algorithm, it counted every spot’s X and Y coordinates in every ROI, and also the center of the spots. The distances between every spot to the center are calculated, and then the mean and the variance of the distances are calculated too. The ratio of variance and mean can be used to discriminate ROI.2) This paper studied two discrimination algorithms, they were Extended Fractal (EF) algorithm and Peak Power Ratio (PPR) algorithm. The EF algorithm can easily eliminate most of the false alarms, but it takes lots of time. However, the PPR algorithm can quickly eliminate the false alarms made by nature things. The combination of the two Feature extraction algorithms can be better.3) This paper particular introduced a software for SAR-ATR, it contained the background and realization of the software. There were several steps in this software, they were de-speckle, detection, clustering, range of interest (ROI) segmentation, discrimination, and classing. The de-speckle processor is a soft threshold wavelet method. An improved constant false alarm rate (CFAR) method is used as the detector. The strong scattering points including to the same ROI are clustered by a desity-based clusterring algorithm named as DBSCAN and its deformation. Maximum a posterior-anisotropic diffusion (MAP-AD) segmentation algorithm makes the ROIs be three parts:target, background and shadow. The segmentation results are used in different stage of the ATR system. And the algorithms mentioned by this paper were used in this software system.