Supervised feature selection for polarimetric SAR classification

Lots of SAR polarimetric features have been proposed to discriminate the different scattering processes of earth terrain. Using the full set of these features for classification is computationally too expensive and some of the features may be irrelevant to the classification task and other may be redundant. Thus, it is useful to exploit the discriminative power offered by a selection and combination of these features. Due to the resulting redundancy and the added computation complexity, an improved sparse support vector machine feature selection algorithm is presented to select a set of discriminative features for efficiently classifying crops by polarimetric SAR. We modify the original algorithm with a simple voting strategy, which extends the original binary-class problem into a multi-class issue. Meanwhile, it can automatically select a feature subset that is well suited for all classes. Experimental results show that the proposed feature selection algorithm can effectively select a good subset of features to discriminate different crops in polarimetric SAR images.