Research on mixed PCA/ICA for SAR image feature extraction

The differences between Principal Component Analysis (PCA) and Independent Component Analysis (ICA) for feature extraction are analyzed theoretically and experimentally, and a mixed PCA/ICA transform is developed for Synthetic Aperture Radar image feature extraction. This method combines the subspace produced by PCA and the subspace generated by ICA to form a mixed subspace to be used to extract features. The mixed components features retain the information characterized by statistics of second and high orders simultaneously. Finally, combined with Support Vector Machine (SVM), the method is employed to recognition of objects in MSTAR SAR dataset. Experimental results indicate the method can improve the recognition performance slightly compared to PCA and ICA.