摘要
针对印刷电路板表面缺陷检测中面板线路复杂、缺陷微小且检测精度与效率难以兼顾的问题,提出一种轻量化、高效的YOLOv8n-LSCNet目标检测模型。首先,在YOLOv8n模型基础上,引入C2f-OREPA模块,利用在线重参数化技术提升特征提取能力;其次,设计一种轻量化检测头,通过共享卷积减少冗余计算;最后,采用扩展交并比(EIoU)损失函数优化边界框回归精度。使用北大印刷电路板(PCB)数据集进行训练与测试,通过消融实验与对比实验验证各模块的有效性。结果表明:相比YOLOv8n模型,YOLOv8n-LSCNet模型的精确率与均值平均精度(交并比阈值≥0.50)分别提升了0.94%和0.47%,参数量与浮点计算量分别降低了21.4%和19.7%;该模型在精度与效率之间取得了良好平衡,具备较强的工程应用潜力。
To address the problems of complex panel circuits,small defects,and difficulty in balancing detection accuracy and efficiency in surface defect detection of printed circuit boards,a lightweight and efficient YOLOv8n-LSCNet object detection model is proposed.First,based on the YOLOv8n model,a C2f-OREPA module is introduced to enhance feature extraction capability utilizing online re-parameterization techniques.Second,a lightweight detection head is designed to reduce redundant computations through shared convolution operations.Finally,an extended intersection over union(EIoU)loss function is adopted to optimize bounding box regression accuracy.The model is trained and tested on the Peking University printed circuit board(PCB)dataset,and both ablation and comparative experiments are conducted to verify the effectiveness of each module.The results show that compared to the YOLOv8n model,the YOLOv8n-LSCNet model improves preision and mean average accuracy(intersection over union threshold≥0.50)by 0.94%and 0.47%,respectively,while reducing parameters and floating-point operations by 21.4%and 19.7%.The proposed model achieves a well-balanced trade-off between accuracy and efficiency,demonstrating strong potential for engineering applications.
作者
赖俊杰
曾猛杰
任洪亮
LAI Junjie;ZENG Mengjie;REN Hongliang(College of Information Science and Engineering,Huaqiao University,Xiamen 361021,China;Fujian Key Laboratory of Light Propagation and Transformation,Huaqiao University,Xiamen 361021,China)
出处
《华侨大学学报(自然科学版)》
2026年第1期61-67,共7页
Journal of Huaqiao University(Natural Science)
基金
福建省厦门市高校科研院所产学研项目(2023CXY0212)。
