针对当前印刷电路板（printed circuit board, PCB）裸板缺陷检测算法对小目标检测准确率较低、误检率过高等问题，提出一种改进的YOLO-PCB缺陷检测算法。该算法在YOLOv5s算法的基础上引入注意力机制，增强特征图的通道特征；同时引入加权双向特征金字塔网络改进特征融合层，使网络实现更高层次的特征融合；而且增加小目标检测层，提高网络对印刷电路板上小目标缺陷的检测能力。实验结果表明，相较于原YOLOv5算法，改进后的检测算法具有更强的特征提取融合能力和更高的检测精度，YOLO-PCB算法的mAP0.5提升了4.08%，mAP0.5：0.95提升了56.69%，精确度提升了1.81%，召回率提升了6.76%。

To solve the problems of low accuracy and high false detection rate of current defect detection algorithms for printed circuit board (PCB) bare board, an improved YOLO-PCB defect detection algorithm was proposed. Based on YOLOv5s algorithm, the new algorithm integrates attention mechanism to enhance the channel features of feature map. At the same time, the feature fusion layer was improved by introducing the weighted bidirectional feature pyramid network, which enabled the network to achieve higher-level feature fusion. In addition, the new algorithm added a small target detection layer to improve the network's detection capability for small target defects on the printed circuit board. The experimental results show that compared with the original YOLOv5 algorithm, the YOLO-PCB detection algorithm has stronger feature extraction fusion capability and higher detection precision, and the new algorithm improves the mAP0.5 by 4.08%, the mAP0.5:0.95 by 56.69%, the precision by 1.81%, and the recall by 6.76%.