From Pixel to Superpixel: A Multi-Scale Strategy for Polarimetric Sar Image Classification

Polarimetric synthetic aperture radar (PolSAR) image classification is a hot topic in remote sensing. However, it is a dilemma that how to preserve image details in PolSAR image meanwhile hinder speckle noise. In this paper, a novel multi-scale strategy is proposed to combine pixel-wise and superpixel-wise features. Specifically, pixel-based polarimetric features and multi-scale superpixel-based polarimetric features are extracted firstly. Then, the multiple feature-induced kernels are fused to form one composite kernel. Finally, the composite kernel is incorporated with a support vector machine (SVM) for terrain classification. Experimental results tested on a real PolSAR image demonstrate the effectiveness of the proposed multi-scale scheme.