基于改进U-Net网络的PCB缺陷检测方法

彭勇1; 刘慧民2; 李伟松1; 王石3

基于改进U-Net网络的PCB缺陷检测方法

引用本文：彭勇1，刘慧民2，李伟松1，王石3.基于改进U-Net网络的PCB缺陷检测方法[J].计算技术与自动化,2025,(1):183-188

摘要点击次数: 1110

全文下载次数: 231

作者	单位
彭勇1，刘慧民2，李伟松1，王石3	（1.东莞职业技术学院电子信息学院,广东东莞 523808；2.东莞森玛仕格里菲电路有限公司,广东东莞 523800；3.东莞华信智能技术有限公司,广东东莞 523808）

中文摘要:针对PCB表面小尺寸缺陷难以检测的问题，提出了一种改进的U-Net 语义分割网络，实现PCB表面缺陷图像的精确检测。首先，将U-Net的四层网络层次修改为三层，可以减少整体的计算工作量、提升网络模型收敛速度、缩短训练时间；其次，在 U-Net 网络中融入CBAM（Convolutional Block Attention Module）模块来提升图像中缺陷目标的显著度；然后，在编码阶段使用混合空洞卷积替换原有卷积块，增大感受野，获取更多的上下文信息。结果表明，U-Net的改进模型能够在提升模型性能的同时减少计算复杂度，能够增加PCB缺陷检测效率。

中文关键词:缺陷检测 U-Net 空洞卷积注意力机制语义分割网络轻量型网络深度学习小目标检测

PCB Defect Detection Method Based on Improved U-Net Network

Abstract:Aiming at the low detection efficiency of the small-size defects of PCB surface, which cannot meet the real time detection requirements industrial production, a defect recognition method based on improved U-Net for printed circuit board defect detection is proposed. Firstly, reduce the original four layers of U-Net to three layers to reduce network computation and shorten model training time. Secondly, the convolutional block attention module （CBAM） was integrated into the U-Net network to improve the significance of the defective targets in the image. Finally ,in the encoding stage, the hybrid dilated convolution is used to replace the original convolution block to increase the receptive field and obtain more context information. The results show that the improved U-Net model network improves network performance while reducing the computational complexity of the U-Net network and can increase the efficiency of printed circuit board defect detection.

keywords:defect detection U-Net dilated convolution attention mechanism semantic segmentation network lightweight network deep learning small target detection

查看全文 查看/发表评论 下载pdf阅读器