Mechanical Cylindrical Part Surface Defect Detection of Components using YOLOv5 and Attention Mechanism

This is the authors' abstract. We don't add key points for this paper.

Effective part defect detection is vital in the manufacturing industry to ensure high-quality standards and minimize the impact of defective parts. Traditional methods lack objectivity and struggle to capture intricate defect patterns, while deep learning approaches face challenges in handling small-sized defects and fusing multi-scale features. To address these limitations, this paper proposes a novel part defect detection network that combines YOLOv5 with attention mechanisms. The network introduces a new backbone network (SKBnLrNet) and a neck network (SAConNet) to enhance feature representation and fusion. The experimental results prove the superiority of the network. The accuracy rate of 50 unqualified parts detection is 100%, and the accuracy rate of 50 qualified parts detection is 98%. This research is helpful to improve the detection level of parts defects.