SAR Target Recognition via Sparsity Preserving Projections

Feature extraction is critical in Synthetic Aperture Radar (SAR) target recognition. Principle Component Analysis (PCA) which preserves global structure and Locality Preserving Projections (LPP) which captures local structure are two typical feature extraction methods in SAR target recognition. But they both keep only one kind of space structure. To combine these two structures, a method of SAR target recognition via Sparsity Preserving Projections (SPP) is proposed in this paper. First, SPP is employed to extract features. It preserves sparse reconstruction information which contains both global and local structure. Natural discriminative information is also kept in sparse reconstruction coefficients without prior knowledge. Then, Sparse Representation based Classification (SRC) is utilized in classification because of its robustness to noise. Experimental results on MSTAR datasets demonstrate effectiveness of our method.