Steel Surface Defect Detection Based on an Improved YOLOv5 Model

Surface defect detection is critical in the process of steel manufacturing. To address the issues of low detection accuracy, slow speed, and low level of intelligence in the detection of steel surface defects, this paper proposes a steel surface defect detection algorithm based on the improved YOLOv5 model. Firstly, the OSA-C3 (One-Shot Aggregation cross stage partial network with 3 convolutions) module is proposed to replace the C3 module in the backbone network, which enhances the model's feature extraction capability through one-shot aggregation. Secondly, The BiC-PAN (Bidirectional Cascade Path Aggregation Network) structure significantly enhances feature fusion with little parameter increase. Finally, the Coordinate Attention (CA) mechanism is introduced into the network to further improve the model's detection performance. Experiments demonstrate that the mean average precision (mAP) of the proposed algorithm is 79.1% on the NEU-DET dataset. Additionally, the average detection speed of the proposed model is 95 frame per second, which balances the detection accuracy and detection efficiency well.