Research on coal gangue recognition method based on XBS-YOLOv5s

Abstract Aiming at the problems of misdetection, omission and low recognition accuracy of coal gangue recognition due to the harsh environmental factors such as low illumination, motion blur and large quantities of coal gangue mixing in coal mines, a coal gangue recognition method based on XBS-YOLOv5s is proposed. Simulate the actual underground production environment to build a machine vision platform, construct a coal gangue image data set, and provide a test environment for various target detection algorithms. In this paper, we construct a real-time detection model of coal gangue in the complex environment of coal mine by fusing SimAM parameter-free attention mechanism, BiFPN feature fusion network and XIoU loss function in YOLOv5s, so as to improve the model’s ability of extracting, fusing and localizing key features of the target. The experimental results show that the recognition accuracy of XBS-YOLOv5s algorithm for coal gangue in the complex environment of low illumination, motion blur and large quantities of coal gangue mixed are effectively improved. Its mean average precision reaches 96%, which is 4.3% higher than the original YOLOv5s algorithm, meanwhile, compared with other YOLO series algorithms, it has the best comprehensive detection performance, which can provide technical support for intelligent and efficient sorting of coal gangue.