Research on lightweight inspection system of steel surface defects based on deep learning model

Addressing the dual challenges of high computational complexity and insufficient detection accuracy faced by existing steel surface defect detection models in industrial deployment, a lightweight detection model based on YOLOv5s is proposed. The lightweight model lays a lightweight foundation with MobileNetV3 as the backbone network and innovatively incorporates the Coordinate Attention (CA) mechanism to enhance spatial localization ability, thereby achieving an optimized balance between detection accuracy and inference efficiency overall. Experiments on the NEU-DET dataset demonstrate that, while maintaining high inference speed, the average precision of the proposed model reaches 91.3%, representing an improvement of 2.1% over the original YOLOv5s baseline model. Furthermore, to verify its generalization capability, cross-component experiments conducted on the bearing defect dataset show that the proposed model attains a detection accuracy of 82.9%, which is 1.4% higher than that of the YOLOv5s baseline with 81.5%, and significantly outperforms other comparative models. This study consequently offers a high-precision, high-efficiency, and well-generalizable detection solution for resource-constrained industrial scenarios.