Abstract: Aiming at the problems of polarimetric synthetic aperture radar (PolSAR) image water and land segmentation with non-smooth and complete edges, and poor segmentation results in small water areas, this paper proposes a PolSAR image segmentation method based on multi-feature support vector machines (SVM) algorithm. Firstly, seven polarization scattering features of a pixel are extracted by Cloude and Yamaguchi target decomposition method, and eight-dimension feature vector of the pixel is constructed by combining with its Sentinel-1 dual-polarized water index (SDWI). All the pixels of the PolSAR image are traversed and obtain the feature set of the PolSAR image. Then, the training dataset are constituted by the feature and class sets of the training samples to train the SVM classifier, which is utilized to realize the water and land segmentation for PolSAR images. Finally, the water and land segmentation experiments of PolSAR images are carried out using the proposed and comparison methods. The experimental results show that the proposed method can realize the water and land segmentation better, the mean values of Kappa coefficients and total accuracy are 0.979 3 and 98.98%, respectively.

Key words: PolSAR image, target decomposition, SDWI, SVM, water-land segmentation