Abstract: The image classification of coal and gangue is the key to realize the automatic separation of coal and gangue images. To improve the accuracy and stability of coal gangue automatic separation model, a hybrid feature selection and classification method based on Relief, MRMR algorithm and SVM classifier is proposed in this paper. A total of 26 features of the color and texture of coal gangue image are used to study its classification. LBP local features and GLCM global features are combined to improve the accuracy of coal-gangue classification. After an optimal feature subset is selected, PSO and SVM are used to construct PSO-SVM optimal parameter model for coal gangue classification. The results show that this method can eliminate more redundant features and improve the efficiency of coal-gangue classification. The average classification accuracy of the model is 96.12% and 94.17% respectively on two different data sets, which proves the effectiveness of the method and the stability of the model.