Feature evaluation and selection for polarimetric SAR image classification

This paper presents an evaluation of different features for polarimetric SAR (PolSAR) image classification. Firstly, we select several of the polarimetric features to give a summary on them. Then we give an insight into their classification performance together with a texture feature using the support vector machine (SVM). Finally, we employ a feature combination and selection strategy that optimizes the trade-off between the feature dimension and precision. The experimental results on PolSAR data of the CETC38 demonstrate: i) the strategy works effectively in the reduction of redundant feature dimensions; ii) in comparison with the unselected feature, the classification performance and computation efficiency of the selected one are improved by this approach.