PCEB-YOLOV8: An Improved PCB Defect Detection Method Based on YOLOV8

Printed circuit board defect recognition is of great significance to ensuring the quality of electronic products. This paper proposes an improved PCB defect detection method based on YOLOV8. Firstly, the data augmentation method PredMix (prediction mix) is adopted to simulate scenarios such as overlapping and occlusion in images, where the displays are incomplete or blurred, thereby enhancing the robustness of the network model. Secondly, BiFPN is used to replace PAN-FPN in the fusion layer to strengthen the feature fusion capability. Finally, the CESE_C2f module is employed in the Backbone part, leveraging a lightweight attention mechanism to improve the detection effect while reducing the number of parameters. Experimental results show that compared with the original YOLOV8, the MAP value of this improved algorithm reaches $\mathbf{92.6}\%$, with a $\mathbf{2.3\%}$ increase in accuracy.