SAR target feature extraction and recognition based on generalized 2DPCA

A feature extraction method based Generalized 2-dimensional Principal Component Analysis (G2DPCA) is presented for Synthetic Aperture Radar (SAR) images. As opposed to PCA, G2DPCA directly seeks the optimal projective axes based on 2D image matrices rather than 1D image vectors, so image matrices do not need to be transformed previously into image vectors. In contrast to 2DPCA, G2DPCA eliminates the correlations of images rows and columns simultaneously. Experimental results based the Moving and Stationary Target Acquisition and Recognition (MSTAR) data show that G2DPCA combining the SAR image preprocessing not only decreases feature dimensions sharply, but increases the correct recognition rates, more than 97%, and is robust to the variation of target azimuth.