SAR Target Recognition with Discriminant Feature Extraction and Hypergraph

Due to the high resolution and multipolarization of Synthetic Aperture Radar (SAR), dimensionality reduction is often needed as a pre-processing step for SAR target recognition. For this purpose, Laplacian Eigenmaps (LE)can be naturally regarded as a candidate, but it neglects the difference between the intra-class and inter-class discriminant information. In this paper, we attempt to integrate LE with Linear Discriminant Analysis (LDA)in order to reduce features for SAR objects. Owing to the inherent imaging mechanism and speckle noise of SAR images, we also propose applying hypergraph to formalize the complex relationships between SAR objects. We have performed extensive experiments on the Moving and Stationary Target Acquisition and Recognition database. Experimental results demonstrate the effectiveness of the presented combination of LE, LDA, and hypergraph on SAR target recognition.