基于YOLOv13的印刷电路板缺陷检测研究

Research on Defect Detection of Printed Circuit Boards Based on YOLOv13

摘要: 印刷电路板(PCB)作为电子设备的核心载体，其表面缺陷直接影响电子产品可靠性与使用寿命。针对传统检测方法效率低、漏检率高、抗干扰能力弱的问题，深度学习目标检测算法已成为PCB缺陷检测的主流技术路径。本文以YOLOv13模型为研究对象，以上海交通大学公开的DeepPCB数据集为实验载体，开展PCB缺陷检测研究。研究设置适配PCB小目标特性的训练参数，实现YOLOv13的端到端训练。针对PCB缺陷特征改进YOLOv13算法，新增缺陷感知注意力模块以强化薄弱类缺陷特征表达，实验结果表明，改进的YOLOv13在DeepPCB数据集上可实现98.7%的mAP50与77帧/秒的检测速度，既能满足工业产线的实时性和精度需求，也能为后续YOLOv13的改进研究提供基础性能基准。

Abstract: As the core carrier of electronic devices, the surface defects of Printed Circuit Boards (PCBs) directly affect the reliability and service life of electronic products. To address the issues of low efficiency, high missed detection rate, and weak anti-interference ability of traditional detection methods, deep learning-based object detection algorithms have become the mainstream technical approach for PCB defect detection. This study takes the YOLOv13 model as the research object and the publicly available DeepPCB dataset from Shanghai Jiao Tong University as the experimental platform to conduct systematic training and testing research on PCB defect detection. Training parameters adapted to the characteristics of small PCB targets are set to realize the end-to-end training of YOLOv13. Targeting the characteristics of PCB defects, the YOLOv13 algorithm is improved by adding a new Defect-Aware Attention (DAA) module to strengthen the feature representation of weak defect categories. Experimental results show that the improved YOLOv13 can achieve an mAP50 of 98.7% and a detection speed of 77 frames per second (FPS) on the DeepPCB dataset. It not only meets the real-time and accuracy requirements of industrial production lines but also provides a basic performance benchmark for subsequent research on the improvement of YOLOv13.