SAR Image Classification Based on Immune Clonal Feature Selection

Abstract

Texture provides valuable information for synthetic aperture radar (SAR) image classification, especially when the single-band and single-polarized SAR is concerned. Three texture feature extraction methods including the gray-level co-occurrence matrix; the gray-gradient co-occurrence matrix and the energy measures of the undecimated wavelet decomposition are introduced to represent the textural information of SAR image. However, the simple combination of these features with each other is usually not suitable for SAR image classification due to the resulting redundancy and the additive computation complexity. Based on immune clonal selection algorithm, a new feature selection approach characterized by rapid convergence to global optimal solution is proposed and applied to find the optimal feature subset. Based on the features selected, SVMs are used to classify the land covers in SAR images. The effectiveness of feature subset selected and the validity of the proposed method are well verified by the experiment results.