摘要
针对PCB线路板缺陷检测问题,采用了卷积神经网络中的新颖架构,即改进的YOLOv5模型,通过对大量带有标记的PCB缺陷图像进行训练,建立了一个高效的缺陷检测网络。在模型的设计过程中,着重考虑了图像中噪声的随机性及PCB生产过程中可能出现的几何变形,引入注意力机制SENet,并改进损失函数。该方法提高了缺陷检测准确率,有效解决了传统方法中漏检和误检问题,为其他电子组件的自动化检测奠定理论基础与技术保障。
In order to solve the problem of PCB circuit board defect detection,this paper adopts a novel architecture in convolutional neural network,i.e.,the improved YOLOv5 model,and establishes an efficient defect detection network by training a large number of PCB defect images with labels.In the design process of the model,the randomness of the noise in the image and the various geometric deformations that may occur during the PCB production process are emphatically considered,the attention mechanism SENet is introduced,and the loss function is improved.This method improves the accuracy of defect detection,which lays a theoretical foundation and technical guarantee for the automatic detection of other electronic components.
作者
张梦婉
周攀
孙权
时睿昕
耿越
ZHANG Mengwan;ZHOU Pan;SUN Quan;SHI Ruixin;GENG Yue(School of Automation,Nanjing Institute of Technology,Nanjing 211167,China;Jiangsu Biomimetic Control Technology and Equipment Engineering Research Center,Nanjing 211167,China)
出处
《电工技术》
2025年第3期96-100,共5页
Electric Engineering
基金
江苏省实践创新训练计划项目(编号202311276073Y)
江苏省配电网智能技术与装备协同创新中心开放基金(编号XTCX201909)。
