Surface defect detection algorithm for copper strip based on SCAP enhancement

Surface defects are an important factor affecting the quality of copper strip products. Deep learning methods can improved the ability of surface defect detection on copper strip, but there are challenges in small object detection, and the algorithm’s generalisation performance is often insufficient. The surface defect detection algorithm that we propose, named SCAP-YOLOv8n, is an enhanced object detection model that combines spatial and attention mechanism with depth coordinated pool to improve the baseline model. Firstly, the ability to small targets detection is improved by introducing a shallow detection scale (160 × 160) in the head network; Secondly, the DS_CARAFE module is used in the neck network to enhance the fusion between features; Thirdly, replacing traditional down-sampling units with HWD modules can better preserve image details; Finally, integrate SCAP modules into the backbone network to improve recognition accuracy. Experiments on the KUST-DET dataset showed that the mAP of the algorithm was 97.58%, the model parameter size was 3.449 MB, and the computational complexity was 12.884 GFLOPs. On the PKU-Market-PCB dataset, the algorithm achieved 92.35% mAP and 95 FPS. Experimental results have shown that SCAP-YOLOv8n has the potential to achieve online detection and good generalisation ability.