Abstract
During the operation of belt conveyors, damage to the conveyor belt (such as tearing, breaking, and scratching) is a major safety hazard. Aiming at the specific challenges of the existing detection methods, such as the difficulty in distinguishing damage from background texture in complex material scenarios, low efficiency of manual inspection, and limited hardware resources, this paper proposes a conveyor belt damage detection algorithm based on the improved YOLOv8. The main improvements include: using the Focal Modulation module to replace the SPPF structure in the backbone network and enhance the feature expression ability; Introduce the dynamic upsampling module (Dysample) , focus on sampling the target area, and effectively suppress the interference of similar backgrounds; Embed the efficient multi-scale attention mechanism (EMA) in the neck network to enhance the model's attention to the injured target; Bounding box regression was performed using the PIoU v2 loss function to optimize the positioning accuracy of irregular damage shapes. The experimental results show that the improved model has achieved a significant performance improvement in the task of conveyor belt damage detection, with an accuracy rate of 90.3% and an average precision mAP of 93.2%. Comparative analysis shows that this method is superior to the original YOLOv8 and traditional detection methods in terms of detection accuracy.