| 825 | 15 | 789 |
| 下载次数 | 被引频次 | 阅读次数 |
针对现有印刷电路板(PCB)缺陷检测算法具有体积大、精度差和漏检率高等问题，提出一种改进的YOLOv5s算法，利用迁移学习进行优化。在主干特征提取网络加入卷积块注意力模块(Convolutional Block Attention Module, CBAM)并引出一个新的蕴含更多丰富细节信息的浅层特征来提高模型信息感知能力，增强网络对小目标的检测。使用C3-CBAM替换加强特征提取网络的C3结构，为了在保障模型检测精度的同时减少模型参数量，使用深度可分离卷积替换下采样。去掉用于大目标的检测头(Yolo Head),避免数据不均衡带来的先验框分配问题。在自制PCB焊锡缺陷数据集上实验表明，改进后算法较原YOLOv5s算法的模型参数量减少24.8%,平均精度均值(mean Average Precision, mAP)达到99.46%,较原YOLOv5s提升了5.45%,证明了改进措施的有效性。
Abstract：For existing PCB defect detection algorithms with large size, poor accuracy and high missing rate problems, an improved YOLOv5s algorithm is proposed and optimized by using migration learning. Firstly, a Convolutional Block Attention Module(CBAM) is added to the backbone feature extraction network, and a new shallow feature with more detailed information is introduced to improve the information perception of the model and enhance the detection of small targets. Secondly, the C3 structure of the enhanced feature extraction network is replaced by C3-CBAM. To reduce the number of model parameters while maintaining model detection accuracy, depth-separable convolution is used to replace down-sampling. Lastly, the detection head(Yolo Head) used for large targets is removed to avoid the problem of a priori frame assignment due to data imbalance. Experiments on a home-made PCB solder defect dataset show that the improved algorithm reduces the number of model parameters by 24.8% compared to the original YOLOv5s algorithm and achieves a mean Average Precision(mAP)of 99.46%, a 5.45% improvement over the original YOLOv5s, demonstrating the effectiveness of the improvement measures.
[1] 李袁.基于改进YOLOv4的目标检测方法研究与应用[D].重庆：重庆邮电大学，2021.
[2] 王恒涛，张上，张朝阳，等.基于YOLOv5的轻量化PCB缺陷检测[J].无线电工程，2022,52(11):2094-2100.
[3] 伍济钢，成远，邵俊，等.基于改进YOLOv4算法的PCB缺陷检测研究[J].仪器仪表学报，2021,42(10):171-178.
[4] HOU Q L,SUN J,HUANG P L.A Novel Algorithm for Tool Wear Online Inspection Based on Machine Vision[J].The International Journal of Advanced Manufacturing Technology,2019,101(9):2415-2423.
[5] 徐衍微.基于机器视觉的PCB表面元器件装配缺陷检测研究[D].南京：南京师范大学，2021.
[6] 王淑青，鲁濠，鲁东林，等.基于轻量化人工神经网络的PCB板缺陷检测[J].仪表技术与传感器，2022(5):98-104.
[7] LI Z Y,YANG Q.System Design for PCB Defects Detection Based on AOI Technology[C]//2011 4th International Congress on Image and Signal Processing.Shanghai:IEEE,2011:1988-1991.
[8] KAUR B,KAUR G,KAUR A.Detection and Classification of Printed Circuit Board Defects Using Image Subtraction Method[C]//2014 Recent Advances in Engineering and Computational Sciences (RAECS).Chandigarh:IEEE,2014:1-5.
[9] LOPEZ-ESCOGIDO D,LUCA A D.2-D High Precision Laser Sensor for Detecting Small Defects in PCBs[C]//2012 9th International Conference on Electrical Engineering,Computing Science and Automatic Control (CCE).Mexico City:IEEE,2012,9:1-6.
[10] 陈辉东，丁小燕，刘艳霞.基于深度学习的目标检测算法综述[J].北京联合大学学报，2021,35(3):39-46.
[11] PARK J M,YOO Y H,KIM U H,et al.D3PointNet:Dual-level Defect Detection Pointnet for Solder Paste Printer in Surface Mount Technology[J].IEEE Access,2020,8:140310-140322.
[12] PARK J Y,HWANG Y,LEE D,et al.MarsNet:Multi-label Classification Network for Images of Various Sizes[J].IEEE Access,2020,8:21832-21846.
[13] HU B,WANG J H.Detection of PCB Surface Defects with Improved Faster-RCNN and Feature Pyramid Network[J].IEEE Access,2020,8:108335-108345.
[14] 胡珊珊，肖勇，王保帅，等.基于深度学习的PCB缺陷检测研究[J].电测与仪表，2021,58(3):139-145.
[15] ZHANG Q F,LIU H R.Multi-scale Defect Detection of Printed Circuit Board Based on Feature Pyramid Network[C]//2021 IEEE International Conference on Artificial Intelligence and Computer Applications (ICAICA).Dalian:IEEE,2021:911-914.
[16] 高伟，周宸，郭谋发.基于改进YOLOv4及SR-GAN的绝缘子缺陷辨识研究[J].电机与控制学报，2021,25(11):93-104.
[17] WOO S,PARK J,LEE J Y,et al.CBAM:Convolutional Block Attention Module[C]//European Conference on Computer Vision.Mubich:Springer,2018:3-19.
基本信息:
中图分类号:TP391.41;TN41;TP18
引用信息:
[1]卢小康,欧阳华兵,陈田,等.基于深度学习的PCB焊锡缺陷检测[J].无线电工程,2024,54(02):276-283.
基金信息:
上海市地方院校能力建设计划项目(22010501000); 上海多向模锻工程技术研究中心资助项目(20DZ2253200); 上海市临港新片区智能制造产业学院资助项目(B1-0299-21-023)~~
2023-05-17
2023-05-17
2023-05-17