Aiming at the image segmentation problem in ship Synthetic Aperture Radar(SAR) image recognition, the method of mathematical statistics is utilized to study the ship SAR image. After analyzing classical K–Means clustering algorithm and Gaussian Mixture Model(GMM), an improved Gaussian mixture model is proposed to segment ship synthetic aperture radar images. The method adopts the Mahalanobis distance to improve classical K–Means method. At the same time, each probability distribution of traditional GMM is further subdivided into individual probability components. In the calculation of auxiliary variables, a gradient ascent algorithm is applied. The experimental results show that the segmentation results obtained by this study are more accurate and more stable than the segmentation method using the classic K–Means algorithm and ordinary Gaussian mixture model.