Improving the YOLOv8-Based Algorithm for Surface Defect Detection of Parts

To address issues such as low efficiency and insufficient detection accuracy in part defect detection, this paper proposes an improved YOLOv8n-based defect recognition algorithm. The algorithm enhances the network's detection speed, strengthens the model's ability to learn defect features, and improves detection accuracy. First, the YOLOv8n detection network undergoes channel-weight-based model lightweighting, which improves detection speed while maintaining accuracy. Next, by introducing a variable boundary awareness technique, the model's ability to learn target boundaries is enhanced, resulting in the improved YOLOv8-CBM model. Finally, experiments validate the superiority of the improved model: the precision and mAP@50 of the enhanced model increase by 8% and 7.4%, respectively.