Based on Low-level Feature Enhancement for Mining Conveyor Belt foreign object detection algorithm

The mining conveyor belt is an important equipment for coal transportation. During the transmission process, foreign objects (such as stones, anchor rods, ironware, etc.) are mixed in the coal transported by the conveyor, resulting in safety accidents such as belt derailment and fire at the coal drop opening. Aiming at the problem of low accuracy of foreign object detection algorithm due to weak characterization of elongated objects and poor weak semantic feature extraction, a low-level feature enhancement foreign object detection algorithm based on YOLOv7-tiny is proposed, designated as FE-YOLO. Firstly, to prevent the decline of the accuracy of special foreign objects such as anchor rods and steel bars, deformable convolution is introduced to enhance the network's ability to extract long target objects. Secondly, the Low-Level Feature Enhancement Module (LFEM) is constructed to utilize multi-scale feature fusion to improve the weak semantic features of foreign objects and reduce the probability of missed and wrong detection. Finally, Ghost Shuffle Convolution(GSConv) is introduced in the feature fusion stage to minimize the channel information loss caused by the change of feature map size and dimension scaling, ensure the feature fusion efficiency and reduce the number of parameters. To verify the performance of the method, the test is carried out on the foreign object dataset of the self-built underground conveyor (a total of 6332 images). The results demonstrate that the [email protected] of this algorithm can reach 0.871, the [email protected]:0.95 can reach 0.535, and the detection speed can reach 77.12fps, verifying the validity and superiority of the proposed algorithm.