目的 将基于深度学习的YOLOv5算法应用于PCB裸板的缺陷检测上，以提高检测的准确率。方法 通过增加特征融合通路,将C2、C3、C4层直接与P2、P3、P4层相连，从而减小信息的损耗;引入更浅层的C2、F2、P2特征图以增加图像的细节信息;并且使用注意力机制SE_block，大幅提高原算法的准确率。结果 改进后的网络的平均精度由91.54%提高至97.36%，提高了5.82%，并且对于各类缺陷，算法的检测精度都能保持在90%以上，满足工业的需求。结论 文中的算法提高了检测精度，体现了浅层信息在小目标检测上的作用，验证了多信息融合通路的优势，彰显了注意力机制的优越性，相比于原算法具有一定的优势。

The work aims to apply YOLOv5 algorithm to defect detection of bare PCB, so as to improve detection accuracy. Feature fusion path was added to directly connect layers C2, C3 and C4 with layers P2, P3 and P4, so as to reduce the loss of information. Shallower C2, F2 and P2 feature images were introduced to increase the details of the image. Moreover, the attention mechanism SE_block was used to improve the accuracy of the original algorithm. The average accuracy of the improved network increased from 91.54% to 97.36%, with a growth of 5.82%. For all kinds of defects, the algorithm could keep a detection accuracy above 90%, which met the needs of industry. The proposed algorithm improves the detection accuracy, reflects the role of shallow information in small target detection, verifies the advantages of multi-information fusion pathway, and highlights the advantages of attention mechanism. Compared with the original algorithm, the proposed algorithm has certain advantages.