Defect detection and multi-scale feature fusion of cold rolled strip based on lightweight YOLOv11 algorithm

Introduction With the rapid and intelligent development of the steel manufacturing process, efficient and reliable detection of surface defects in cold-rolled strip steel is of great significance to product quality and production safety. However, existing detection methods have limited ability to identify small defects, and some high-precision models have problems such as large parameter scale and insufficient real-time performance. Methods This study proposes a cold-rolled strip defect detection and MSFF method based on lightweight YOLOv11. This method introduces an MSFF structure at the neck of the detection network and combines it with a SimAM-Enhanced Block to enhance the expression of key defect features. Results Experiments demonstrate that the designed method achieves an mAP of 57.85% in the cold-rolled strip defect detection task (IoU threshold 0.50-0.95), which is 4.67% and 11.61% higher than YOLOv8n’s 53.18% and YOLOv5s’ 46.24%. It significantly reduces the risk of missed detection while maintaining high detection accuracy, and the recall rate reaches 88.12%, which is significantly higher than YOLOv8n’s 83.24%, indicating that it has stronger detection capabilities for small and low-contrast defects. Discussion In summary, the proposed method achieves a good balance between detection accuracy, robustness, and engineering deployability, and can provide an efficient and practical solution for online defect detection of cold-rolled strip steel.