The invention belongs to the technical field of radar signal processing, discloses a SAR image matching method based on improved SIFT, and mainly solves the problem that when the signal-to-noise ratioof a SAR image is low and a scene difference exists between the SAR image and a reference image, the conventional matching algorithm cannot satisfy the accurate matching requirement. The realizationsteps of the method include: performing Gamma_MAP filtering on the image, and suppressing the influence of speckle noise; then constructing a difference of Gaussian scale space, detecting extreme points in the scale space, screening and eliminating the extreme points, obtaining characteristic points, and further obtaining a principal direction and a characteristic vector of each characteristic point; and finally obtaining image matching points according to a bidirectional Euclidean distance criterion. According to the method, the stability of the image characteristic points and the matching accuracy are improved, high-precision matching pairs can be obtained, and the method can be used for matching of the SAR images with low signal-to-noise ratio.