In this paper, the traditional FCM algorithm membership function was improved. The improved algorithm can reduce the isolation point of the image data clustering results. In this paper, Gray-gradient co-occurrence matrix of the image texture feature extraction using principal component analysis on the extracted high-dimensional feature image to reduce the dimensions, combined with this improved FCM clustering algorithm to the image after the image data preprocessing clustering. Experiments show that the method has better clustering results, with fewer iterations and can reach the global optimum.