Coal and gangue recognition under four operating conditions by using image analysis and Relief-SVM

Coal and gangue recognition is a key issue in rock picking during the coal preparation process. Four operating conditions including raw coal with the dry clean surface, wet clean surface, dry surface covered by slime and wet surface covered by slime are frequently encountered in real applications. The Relief-SVM method was presented in this paper to classify coal and gangue under these different surface conditions. 19 features including the color features and textural features were extracted first. The relief-SVM method was employed to find the optimal features and build the best classifiers. It was proved to be valid by the experiments of both Dafeng coal and Baijigou coal. Fewer optimal features had the same classification ability as the original features. The mean accuracy ranged from 95.5% to 97% and 94% to 98% for Dafeng coal and Baijigou coal respectively, which showed that the proposed method was suitable for different operating conditions.