Feature Extraction Based on 2D Compressive Sensing for SAR Automatic Target Recognition

This paper proposes a new feature extraction method for synthetic aperture radar (SAR) images with application to automatic target recognition (ATR). The original SAR image is first represented by a sparse image containing only a few dominant scattering centers (SCs). According to the theory of 2D compressive sensing (CS), a sparse image can be reconstructed from a low dimensional matrix with little distortion. Therefore, we use 2D random projection to extract features from the sparse image. The proposed method directly works on the 2D images thus avoiding the conversion of 2D matrices to vectors. Based on the extracted feature, the nearest neighbor classifier (NNC) is employed for target recognition. Experiments are conducted on the moving and stationary target acquisition and recognition (MSTAR) to evaluate the validity of the proposed method. Comparison with other methods demonstrates the superiority of the proposed method.