BA-YOLO: A Lightweight Multi-scale Detection Model for Longitudinal Tear of Conveyor Belts

Abstract

To enhance the multi-scale adaptability and real-time performance of the longitudinal tear detection network for conveyor belts, this paper proposes a lightweight tear detection algorithm based on laser line assistance, which is an improved version of the YOLOv11n model. Firstly, a bidirectional feature pyramid network (BiFPN) is introduced into the neck network to strengthen the network's ability to fuse features of different scales, enabling it to detect tear faults of different sizes on conveyor belts; Secondly, the downsample convolution layer of the main backbone network is replaced by an ADown layer, which reduces the size of the feature map and computational cost through a more efficient feature compression method, thereby effectively reducing the model complexity. Finally, the network is trained on the conveyor belt tear fault dataset. Experimental results show that the precision and recall rate of the BA-YOLO model are 92.4% and 82.8% respectively, and the mean average precision (mAP@0.5) is 90.3%. Compared with the original model, the model parameters have been reduced by 38.9% and the computational cost has been reduced by 14.3%. The improved model not only demonstrates the best detection performance but also shows stronger adaptability.