Classification of coal and gangue under multiple surface conditions via machine vision and relief-SVM

Coal and gangue classification is a key problem in automating the task of gangue picking during the preparation of coal. Four coal properties, namely raw coal with a dry clean surface, a wet clean surface, a dry surface covered by slime, and a wet surface covered by slime, are frequently encountered in real situations. Typically, the conditions occur simultaneously to yield multiple surface conditions. In the study, the relief-Support Vector Machine (relief-SVM) method is proposed to recognize coal and gangue based on image analysis. First, 19 features of coal and gangue pictures including color and textural features were extracted. Subsequently, the relief-SVM method was employed to identify optimal features and construct optimal classifiers. The classifiers were then used on coal samples from the Dafeng and Baijigou coal mines to validate their efficacy in recognizing coal and gangue. The average accuracy corresponded to 92.57% and 92% for Dafeng coal and Baijigou coal, respectively. The experimental results indicated that fewer optimal features increased the classification accuracy and decreased the training and classification time, and thus, the proposed method is suitable for complex conditions. (C) 2019 Elsevier B.V. All rights reserved.