Surface defect detection of steel based on DAC-YOLO algorithm

In industrial settings, current steel surface defect detection technology faces problems such as low detection accuracy and high model complexity. To address these issues, this paper proposes a new DAC-YOLO algorithm. Firstly, introduce deformable convolution (C2f Deformable Convolution, C2f-DCN) into the backbone network to adaptively change the shape and position of the convolution kernel; Secondly, introducing the Asymptomatic Feature Pyramid Network (AFPN) into the neck trunk network improves the model's recognition performance for multi-scale defects; Finally, cross layer connections were added to the PAN-FPN structure to improve the utilization of original features in the backbone network layer. The experimental results on the NEU-DET dataset show that the average accuracy of the DAC-YOLO algorithm reaches 80.0%. Compared with other common steel surface defect detection algorithms, this algorithm has higher accuracy and lower model complexity, which can better meet the requirements of practical industrial production.