DSC-YOLOv11: An efficient method for detecting industrial surface defects

The automated detection of industrial product surface flaws is critical for sustaining product excellence and guaranteeing operational safety. Existing one-stage object detection algorithms exhibit significant limitations when applied to industrial surface defect inspection, including low accuracy, high false alarm rates, and missed detections. So this paper proposes the DSC-YOLOv11 algorithm to enhance the ability of industrial surface defect detection. Firstly, the DD – CBS module is incorporated into the Backbone to address the loss of key information caused solely by downsampling through the Conv module. Secondly, we introduce the SCA – C3k2 module to enhance the model’s focus on the feature transfer nodes between the Backbone and the Neck, thus improving the detection performance. In conclusion, to correct the deficiencies of the original detection head in feature separation, computational cost, and detection accuracy, we propose the CPC – Detect head. The proposed DSC – YOLOv11 architecture was tested on the GC10 – DET dataset. We compared its performance with that of typical one – stage object detection algorithms. The results indicate that our method outperforms others. To further evaluate the generalizability of DSC – YOLOv11, comprehensive experiments were conducted on the NEU – DET dataset, revealing significant performance improvements compared to baseline methods. The algorithm has been verified to enhance defect detection performance, meeting industrial requirements.