Aiming at the problem of high false detection rate of PCB defect detection, an improved YOLOv8-C2f_DBB-SPFF_LSKA (YOLOv8-CDSL) is used to detect defects of the PCB. The diverse branch block (DBB) is used to improve the faster version of CSP bottle-neck with two convolutions (C2f) module in the YOLOv8 backbone network. The large separable kernel attention (LSKA) mechanism is also added to the SPPF module of the YOLOv8 network. The C2F_DBB module can significantly improve the model’s ability to identify small-scale PCB defects, and greatly enhance the model’s performance in comprehensive feature extraction. Thus significantly improving the overall accuracy of the model. The SPPF_LSKA module can reduce the computing power consumption of model training. Consequently, it significantly enhances the detection capability of the improved YOLOv8-CDSL network. The effectiveness of the improved scheme was verified by ablation experiments. At the same time, it is verified by comparative experiments that the improved YOLOv8-CDSL network has the highest detection accuracy of 99.13% for six common surface defects of PCB.