CCD-YOLO: An Accurate and Fast Algorithm for Steel Surface Defect Detection

Abstract A variety of surface defects can be created in steel during manufacturing and transportation, which can reduce the strength of the steel and shorten the service life of the workpiece. Therefore, surface defect detection is a key link to ensure the quality of industrial production. However, traditional surface defect detection methods have the disadvantages of low accuracy and slow speed. Therefore, we propose a steel surface defect detection model CCD-YOLO based on YOLOv5. Firstly, we replace the C3 module in the backbone of YOLOv5 with the C2f module to obtain more gradient flow information while maintaining lightweight. Secondly, we insert a flexible and lightweight CA attention mechanism into the backbone to help the model accurately locate and identify objects of interest. Finally, we use decoupled heads to separate the regression and classification tasks, which improves detection accuracy. Finally, a large number of experimental results show that CCD-YOLO achieves an accuracy of 72.9% mAP on the NEU-DET dataset, which is 4.3% better than YOLOv5 and 1.1% higher than YOLOv8. The model has good comprehensive performance in steel surface defect detection.