基于深度学习的PCB缺陷检测技术  

PCB defect detection technology based on deep learning

作　　者：程立英[1,2] 张文雅 程强 谷利茹 管文印 张志美 CHENG Liying;ZHANG Wenya;CHENG Qiang;GU Liru;GUAN Wenyin;ZHANG Zhimei(College of Physics Science and Technology,Shenyang Normal University,Shenyang 110034,China;Key Laboratory of Intelligent Computing in Medical Image,Ministry of Education,Shenyang 110003,China)

机构地区：[1]沈阳师范大学物理科学与技术学院,沈阳110034 [2]医学影像智能计算教育部重点实验室,沈阳110003

出　　处：《沈阳师范大学学报（自然科学版）》2024年第2期151-156,共6页Journal of Shenyang Normal University:Natural Science Edition

基　　金：教育部产学合作协同育人项目(230704037121136);辽宁省教育厅科学研究经费项目(LZD202003)。

摘　　要：印刷电路板(printed circuit board,PCB)是电子产品的关键部件。在实际生产过程中,PCB难免会产生多种缺陷,对缺陷进行及时、精准检测具有一定的研究意义与应用价值。传统的检测方法存在速度慢、成本高、精度低的问题。针对PCB缺陷检测问题,开展基于YOLO系列算法研究,在相同的实验环境下,以平均精度、精确率、召回率、每秒传输帧数作为评价性能指标。实验研究发现,YOLOv7在精度方面比YOLOv5有一定的提升,而YOLOv5在训练和推理的速度上比YOLOv7更快。提出融合CBAM(convolutional block attention module)注意力机制模块的YOLOv5改进算法用于PCB缺陷检测。经实验验证,改进算法在PCB缺陷检测的精确性和速度性能上均得到提升,其中,平均精度、精确度和召回率分别提升了7.40%,3.57%和5.63%。Printed circuit board(PCB)is a key component of electronic products.It is important to detect the PCB defect timely and accurately in the production process.There are some problems with the traditional detection methods,such as slow speed,high cost and low accuracy.Aiming at the PCB defect detection problem,a series of experiments were conducted based on YOLO detection algorithm to test the average accuracy,precision,recall,and FPS(frames per second)in the same experimental environment.Experimental results showed that YOLOv7 has a certain improvement in accuracy than YOLOv5,and YOLOv5 is faster than YOLOv7 in training and inference.In order to improve the performance,an improved algorithm based on YOLOv5 was proposed,the network structure of YOLOv5 was fused the convolutional block attention module(CBAM).Experiments verified that the average accuracy is improved by 7.40%,the accuracy and recall rate are also improved by 3.57%and 5.63%than those of theYOLOv5 respectively.

关 键 词：PCB缺陷检测 深度学习 YOLOv5 CBAM注意力机制 

分 类 号：O343.1[理学—固体力学] O341[理学—力学]