An Image Recognition Approach for Coal and Gangue Used in Pick-Up Robot

摘要

Picking gangue from raw coal is a crucial step of coal production. Due to the potential for replacing manual workers, the study of pick-up robot is attracting much interest. Pick-up robots usually work in fixed working areas where the types of coals and gangues are unitary. Based on this fact, this paper proposes a simple, fast, and easily implemented approach for coal and gangue classification which is LS-SVM (Least Square Support Vector Machine) based using gray scale and texture as features. We firstly sampled the image dataset from Han City, Shaanxi province and Jizhong, Hebei province which are two main mining areas in China. The data of Han City consists of the images of lean coal and shale, and the data of Jizhong is coking coal and sandstone. By analyzing the gray scale and the texture of the sampled data, we discover that coal and gangue vary in the parameters including the mean and peak of gray scale, contrast ratio, and entropy. Therefore, these four parameters are chosen as features. We utilize LS-SVM as the machine learning model, and the model is trained with three groups of parameters separately. The first are the mean and peak of gray scale, the second are the contrast ratio and entropy which represents texture features, and the third are the peak of gray scale and the contrast ratio which integrates gray scale and texture features. After evaluation by using our sampled dataset, the model trained by the third group outperforms others. The classification results were 98.7% correct of coal and 96.6% correct of gangue for the data of Han city, and 98.6% correct of coal and 96.6% correct of gangue for the data of Jizhong.