The sorting of coal gangue is of great significance for improving coal quality, avoiding environmental pollution, and reducing labor costs. The image-based coal gangue sorting method has been proposed by a large number of researchers, but the complexity of the environment, the speed and accuracy of coal gangue detection and recognition methods, and the performance of hardware equipment all pose challenges to the accuracy of coal gangue sorting. This paper discusses the research and application of deep-learning methods in the field of coal gangue detection and proposes an improved YOLOv7 coal gangue detection model for ordinary GPU devices with large computing power and memory. In response to the feature redundancy problem of the YOLOv7 model in coal gangue detection tasks, FasterNet was introduced to improve the backbone network of YOLOv7, reducing redundant calculations and memory access, making the model more effective in extracting features. In response to the requirements for detection speed in high-speed motion of belt conveyors, VoVGSCSP was introduced to improve the efficient layer aggregation network (ELAN) of YOLOv7 neck, further enhancing the detection speed of the model. The experimental results show that when the belt speed is 0.6 m/s, the improved model’s mAP0.5 is similar to YOLOv7, FPS increases from 9 frames per second to 18 frames per second, coal gangue sorting rates reach 91.1%, and coal misselection rates are 4.8%. The proposed coal gangue detection and recognition method based on improved YOLOv7 has increased the detection speed of the recognition model and promoted the improvement of coal gangue sorting efficiency.
Hou et al. (Sat,) studied this question.