Integrating Motion Deblurring with Deep Learning for Real-Time Defect Detection in High-Speed Steel Production

In the steel manufacturing industry, the precision of surface defect detection on steel plates is a critical factor for optimizing production quality and operational efficiency. As production line speeds continue to escalate, the resultant motion blur from the rapid movement of steel plates increasingly challenges the performance of imaging systems, thereby diminishing the accuracy of defect detection. This paper presents a novel approach to real-time steel plate surface defect detection, leveraging advanced motion deblurring techniques to mitigate these challenges. Specifically, we compare the efficacy of three methodologies: the standalone YOLOv5 defect detection algorithm, YOLOv5 in conjunction with Deblur-GAN preprocessing, and the YOLO-Steel-GAN framework, which seamlessly integrates Deblur-GAN with YOLOv5. Experimental evaluations reveal that while YOLOv5 alone achieves a recall rate of 91.6% with a detection speed of 62 FPS, its precision is limited to 55.2% and a mean Average Precision (mAP) of 51%. The introduction of Deblur-GAN as a preprocessing step with YOLOv5 enhances precision to 63.1% and recall to 92.3%, albeit with a reduction in detection speed to 45 FPS. In contrast, the proposed YOLO-Steel-GAN framework not only sustains a competitive detection speed of 53 FPS but also significantly elevates precision to 89.3%, recall to 94.4%, and mAP to 87.9%. These results demonstrate that the YOLO-Steel-GAN framework provides a robust and efficient solution for real-time steel plate surface defect detection, offering substantial improvements in both accuracy and processing speed. The findings underscore the practical applicability of this integrated approach in high-speed industrial environments, marking a significant advancement in steel plate quality control.