Marginal sample discriminant embedding for SAR automatic target recognition

Feature extraction is a crucial step in synthetic aperture radar (SAR) automatic target recognition (ATR). In this paper, we propose a feature extraction method named marginal sample discriminant embedding (MSDE) which is based on manifold learning theory. This method can preserve class information and neighborhood information of original data during dimensionality reduction. It keeps neighbor relations of within-class samples and separates between-class samples in the low-dimensional feature space. In this method, sample discriminant coefficient is employed to give marginal sample an extra weight. Due to sample discriminant coefficient, discriminative capability of MSDE is enhanced. Experimental results based on MSTAR database show that the proposed method can improve recognition performance effectively.