A lightweight steel surface defect detection network based on improved YOLOv8s

Steel's strength and versatility make it essential in industries like automobile manufacturing and aerospace. However, steel surface defects often arise during manufacturing, impacting the material's performance. Timely detection and remediation of these defects are critical, yet existing algorithms struggle to balance accuracy with real-time performance. Thus, this paper introduces D-Ghost, a lightweight module designed to enhance the YOLOv8s algorithm. Integrated into YOLOv8s, D-Ghost reduces computational complexity while maintaining high detection accuracy. Experimental results show that, compared to the initial model, YOLOv8s (Ghost) has a 1.5% reduction in mAP, while YOLOv8s (D-Ghost) exhibits only a 0.1% decrease in mAP and maintains an accuracy of 74.1%. The D-Ghost module also improves model weight and GFLOPs performance, offering a more efficient solution for industrial defect detection. This approach effectively compresses the model and reduces computational demands without significantly sacrificing accuracy.