Tensor global and local discriminant embedding for SAR target configuration recognition

Tensor linear discriminant analysis (LDA) is an effective feature extraction method for images, but it just considers the globally discriminative information of the data and neglects to preserve the local structure. In this paper, we propose a feature extraction approach based on tensor globally and locally discriminative information preserving projections for SAR target configuration recognition. We first represent SAR images as second-order tensors, and then use the known aspect angles to construct two local adjacent graphs to represent the local structure because SAR images are very sensitive to aspect angles. Finally an optimization problem is obtained which can be solved with the eigenvalue decomposition method by combining the local structure preservation with tensor LDA. Experiments are carried out on Moving and Stationary Target Acquisition and Recognition (MSTAR) public database to evaluate the performance of the proposed method. Experimental results demonstrate the effectiveness of the proposed method.