Research on defect recognition method of mine belts based on deep learning

The traditional manual inspection method is inefficient and prone to missed detections. Although the existing non-destructive testing technologies such as ultrasonic and infrared imaging equipment can provide accurate results, they are complex to operate and have high equipment costs. This paper proposes a defect recognition method for mining belts based on deep learning, which uses the You Only Look Once (YOLO) object detection algorithm to automatically identify and classify surface cracks, wear, tears and other defects of the belts. By analyzing the network structure and experimental results of the You Only Look Once version 8 (YOLOv8) algorithm, it is found that its recognition accuracy is poor for defect categories with small sample sizes, samples with complex environments, and samples with high noise levels. To solve these problems and improve the robustness and accuracy of the model in complex environments, the attention mechanism and the Focal Loss function were introduced. YOLOv8 is improved by introducing the Convolutional Block Attention Module (CBAM) and Focal Loss. Finally, the effectiveness of the method we proposed was verified through experiments. The experimental results show that the improved model demonstrates better accuracy than the original model in identifying all five types of defects.