Kernel marginal sample discriminant embedding for SAR automatic target recognition

Abstract

Synthetic aperture radar (SAR) has been widely used in remote sensing. Feature extraction is a crucial step in SAR automatic target recognition (ATR). In this paper, Kernel Marginal Sample Discriminant Embedding (KMSDE) is proposed, which is based on kernel trick and manifold learning theory. In feature extraction via KMSDE, the original dataset is mapped to high dimensional space and manifold learning theory is introduced for dimensional reduction. KMSDE preserves local and class information of the original dataset, as well as gathers the with-class samples and separates between-class samples in the low-dimensional space. In addition, it employs information related to each sample's location in the dataset, which enhances the discriminative capability of the method. Compared to other SAR imagery feature extraction methods, the experiments based on MSTAR database show that the proposed method improves the recognition performance.