The invention discloses a SAR image segmentation method based on SAR-KAZE characteristic extraction, and mainly solves the problems that an existing method is low in SAR image segmentation precision and efficiency; the segmentation process comprises the following steps: 1, inputting an original to-be-segmented SAR image, and carrying out double-precision transition for the image; 2, carrying out mirror image expansion and partitioning for the double-precision matrix so as to obtain an image block of the double-precision matrix; 4, extracting KAZE features of the SAR image for the image block,thus obtaining a SAR-KAZE feature matrix X of the double-precision matrix; 5, using an improved sparse subspace cluster to cluster the SAR-KAZE feature matrix X, thus obtaining a clustering result ofthe double-precision matrix. The extracted SAR-KAZE features can keep more image margin and detain information, thus improving the segmentation precision; the improved sparse subspace cluster is employed to cluster the image, thus reducing the segmentation complexity; the method can be applied to ground object target classification in radar imaging.