The imaging process of SAR sea ice image is blurred by random factors, resulting in unclear image, which increases the difficulty of automatic interpretation of SAR sea ice images. In view of the above problems, this paper proposes an automatic classification of SAR sea ice images combined with the Retinex and the Gaussian Mixture Model algorithm (R-gmm). Firstly, the SAR image is convoluted by Gaussian function, then the image is optimized by EM algorithm and GMM model, and finally the output image is obtained. The experimental results show that this algorithm effectively enhances the sharpness of SAR sea ice image and improves the segmentation accuracy of SAR sea ice image, which Promotes the realization of SAR sea ice image interpretation automation to some extent.