基于改进YOLOv5的贴片电感表面缺陷检测研究
Research on surface defect detection of SMD inductors based on improved YOLOv5
陈建春 1乔健 2朱子唯 3王功伟4
作者信息
- 1. 联通(广东)产业互联网有限公司,广东广州 510000
- 2. 佛山大学广东省工业智能检测技术重点实验室,广东佛山 528225
- 3. 西交利物浦大学智能工程学院,江苏苏州 215028
- 4. 季华实验室微波与真空室,广东佛山 528200
- 折叠
摘要
为实现贴片电感表面缺陷的快速精准检测,突破目前贴片电感表面缺陷检测速度慢、准确率低的技术难题,在YOLOv5 算法基础上,引入SE注意力模块和双向特征融合网络(BiFPN)模型,提出基于注意力机制的特征提取网络结构,分别对不同特征通道赋予相应权重信息,使其在特征融合中能够快速传递,进一步提高了贴片电感表面缺陷模型的检测精度;考虑提取网络时无法高效检测出贴片电感的缺陷类型,设计出基于加权双向特征金字塔结构,增强了模型对不同尺度特征信息的表达能力;利用贴片电感表面缺陷检测数据集完成了SE注意力机制和BiFPN网络的消融实验以及目标检测算法的对比实验.结果表明,提出的改进模型平均准确率均值(mAP)达到 97.12%,较原YOLOv5 算法提升了 5.87%,检测速度达到 40.47FPS,能够满足贴片电感表面缺陷检测的实时性和准确性要求.
Abstract
In order to achieve fast and accurate detection of surface defects of SMD inductors and break through the current technical problems of slow speed and low accuracy of surface defect detection of SMD inductors,this paper introduced SE attention module and bidirectional feature pyramid network(BiFPN)model based on YOLOv5 algorithm,proposed a feature extraction network structure based on attention mechanism,and assigned corresponding weight information to different feature channels,so that they can be quickly transferred in feature fusion.The detection accuracy of the surface defect model of SMD inductor is further improved.Considering that the defect type of the SMD inductors cannot be detected efficiently when extracting the network,a bidirectional feature pyramid structure is designed to enhance the ability of the model to express the feature information of different scales.The ablation experiment of SE attention mechanism and BiFPN network,as well as the contrast experiment of target detection algorithm,are completed by using the SMD inductors surface defect detection data set.The results show that the mean average precision(mAP)of the improved model proposed in this paper reached 97.12%,which is 5.87%higher than the original YOLOv5 algorithm,and the detection speed reached 40.47 FPS,which can meet the real-time and accuracy requirements of surface defect detection of SMD inductors.
关键词
缺陷检测/YOLOv5/SE注意力模块/BiFPNKey words
defect detection/YOLOv5/SE attention module/BiFPN引用本文复制引用
基金项目
广东省重点建设学科科研能力提升项目(2022ZDJS042)
珠江人才计划项目(X200221DA200)
出版年
2024
国家科技期刊平台
NSTL
万方数据