辽宁工程技术大学学报（自然科学版）2023，Vol.42Issue(6)：756-762,7.DOI:10.11956/j.issn.1008-0562.2023.06.016
基于深度学习的国产工业相机PCB缺陷检测算法
PCB defect detection algorithm for domestic industrial cameras based on deep learning
摘要
Abstract
In response to the problem of printed circuit board(PCB)defect detection,which imported complete set of systems are expensive,closed-source,and unsupported secondary development,at same time,the efficiency and accuracy of the core detection algorithms for domestic industrial cameras hardware are poor,a PCB defect detection algorithm for domestic industrial cameras based on deep learning was proposed.Firstly,the PCB training sample set was collected by domestic industrial cameras,then generated an anchor box that meets the defect size according to the defect characteristics by the K-means++ algorithm.Secondly,a feature layer with a scale of 104×104 was added to extract more scale feature information based on the YOLOv3 network structure for deep learning algorithm.Finally,the defect location and category were obtained by the joint prediction of multi-scale features.The experimental results show that the mean Average Precision of the proposed algorithm is 97.42%,which better than SSD,YOLOv3 and Faster RCNN algorithm at same level of time cost,and can meet the actual needs of PCB defect detection for domestic industrial cameras.关键词
国产工业相机/PCB缺陷检测/YOLOv3/K-means++/多尺度特征Key words
domestic industrial camera/printed circuit board defect detection/YOLOv3/K-means++/multi-scale features分类
信息技术与安全科学引用本文复制引用
陈万志,阴晓阳,方圆,房娜..基于深度学习的国产工业相机PCB缺陷检测算法[J].辽宁工程技术大学学报（自然科学版）,2023,42(6):756-762,7.基金项目
国家重点研发计划项目(2018YFB1403303) （2018YFB1403303）
国家级大学生创新创业训练计划项目(201710147312) （201710147312）
