Synthetic Aperture Radar Target Feature Extraction and Recognition Based on Improved Sparsity Preserving Projections

Abstract

We have proposed an improved Sparsity Preserving Projection (SPP) method to implement target feature extraction. It combines the SPP feature extraction using the idea of the Locality Preserving Projection (LPP) scheme to build a new objective function, which can not only maintain the relationship of sparse reconstruction between the samples but also minimize the distance between similar sample types in the projection space. Experimental results with Moving and Stationary Target Acquisition and Recognition (MSTAR) Synthetic Aperture Radar (SAR) data sets show that the average recognition rate using the proposed method is up to 97.81% without knowing the target to be azimuth, which can improve the target recognition result even further for obvious reasons. The proposed method is an effective one for SAR target recognition.