Detection of coal content in gangue via image analysis and particle swarm optimization–support vector machine

Abstract

To detect coal content in gangue, a novel approach based on image analysis and particle swarm optimization–support vector machine (PSO-SVM) was presented. First, 15 features that included four sizes and 11 density parameters were extracted from coal and gangue regions of sample pictures, respectively. For the size parameters, the values of each feature are summed up by class, while for the density parameters, the average operation was implemented. Then, the values of coal features were divided by that of gangue features to obtain final features. Using the feature selection method based on the Pearson correlation coefficient, we identified six features that best demonstrated that a consideration of the interaction between size and density parameters can achieve better prediction results. Finally, the coal content in the gangue model was determined using SVM optimized by PSO. The experiment was repeated three times, and the average relative errors were 10.0%, 9.8%, and 9.5%.