2DPCA-based two-dimensional marginal sample discriminant embedding for SAR ATR

Feature extraction is a key step in synthetic aperture radar (SAR) automatic target recognition (ATR). In this paper, we propose a feature extraction algorithm based on manifold learning theory, the algorithm is named Two-dimensional Principal Component Analysis-based Two-dimensional Marginal Sample Discriminant Embedding (2DPCA-based 2DMSDE). Above all, the original SAR images are projected by 2DPCA which is effective for feature representation, the dimension of SAR images is reduced in horizontal direction and global information of the original dataset is preserved. Furthermore, 2DMSDE is employed to reduce dimension in vertical direction , preserve local information of the dataset and enhance discriminative ability. Therefore, 2DPCA-based 2DMSDE not only further compresses the dimensions of original images, but also achieves better recognition performance. Experimental results demonstrate the effectiveness of 2DPCA-based 2DMSDE.