Combining wavelet invariant moments and relevance vector machine for SAR target recognition

A new method to classify targets in SAR images is proposed in this paper. The method combines both the advantages of wavelet invariant moments and relevance vector machine (RVM). The wavelet invariant moments have the wavelet inherent property of multi-resolution analysis and moment invariants quality. We firstly extract wavelet invariant moments to indicate targets in SAR images, and then select the feature set with principal component analysis (PCA). Finally, the selected feature set is fed to RVM for training and classifying. The RVM can powerfully manage complex classification and regression problems basing on the concept of probabilistic Bayesian learning framework. We perform 2- class and 3-class classification experiments respectively. Test results show that wavelet invariant moments indicate the target effectively, and RVM performs better than K-nearest neighbourhood (K-NN), back propagation neural network (BPNN) and least square support vector machine (LSSVM).