In coal production, the raw coal contains a large amount of gangue, which affects the quality of coal and pollutes the environment. Separating coal and gangue can improve coal quality, save energy, and reduce consumption and make rational use of resources. The separated gangue can also be reused. Robots with computer vision technology have become current research hotspots due to simple equipment, are efficient, and create no pollution to the environment. However, the difficulty in identifying coal and gangue is that the difference between coal and gangue is small, and the background and prospects are similar. In addition, due to the irregular shape of gangue, real-time grasping requirements make robot control difficult. This paper presents a coal and gangue separating robot system based on computer vision, proposes a convolutional neural network to extract the classification and location information, and designs a robot multi-objective motion planning algorithm. Through simulation and experimental verification, the accuracy of coal gangue identification reaches 98% under the condition of ensuring real-time performance. The average separating rate reaches 75% on low-, medium-, and high-speed moving conveyor belts, which meets the needs of actual projects. This method has important guiding significance in detection and separation of objects in complex scenes.
Keywords: YOLO; coal and gangue; detection; robot.