Intelligent selection of useful features for optimal feature-based classification

In feature-based classification, each target class is characterised by a reference feature vector that comprises a combination of physical and statistical attributes. Different combinations of features are useful to distinguish amongst different target classes. In this study, an intelligent features selection method is proposed which selects features with minimum intra-class variance/inter-class variance. Classification results obtained with MSTAR data for tanks, APCs and trucks have shown a significant improvement in classification performance over using all measured features