YOLO-DHC:A Multi-scale method for PCB defect detection

The stability of the equipment’s functioning may be impacted by surface flaws on the printed circuit board (PCB) caused by factors like the production environment. Aiming at the issues of small defect target, complex defect shape, missed detection and false detection in PCB surface defect detection, a YOLO-DHC surface defect detection algorithm based on improved YOLOv8 model is proposed. Firstly, the C2F-DC module based on dynamic convolution is proposed to replace the C2f module in the backbone network, which improves the feature extraction ability of the model by adaptive generation of convolutional kernel weights. Secondly, the HSD module is introduced to retain the details of defects in the process of downsampling. Finally, the CMSF multi-scale fusion module is presented to enhance the model’s detection capabilities even more. Compared with YOLOv8s, the experimental results show that the average accuracy (mAP) of the algorithm on the public PCB dataset is increased by 2.4%, and the recall rate is increased by 3.1%.