基于融合MBAM与YOLOv5的PCB缺陷检测方法

doi:10.11996/JG.j.2095-302X.2024010047

图学学报 ›› 2024, Vol. 45 ›› Issue (1): 47-55.DOI: 10.11996/JG.j.2095-302X.2024010047

• 图像处理与计算机视觉 • 上一篇下一篇

基于融合MBAM与YOLOv5的PCB缺陷检测方法

胡欣¹(), 胡帅¹, 马丽军², 司利云¹(), 肖剑³, 袁晔⁴

1.长安大学能源与电气工程学院，陕西西安 710064
2.国网甘肃省电力公司天水电力公司，甘肃天水 741000
3.长安大学电子与控制工程学院，陕西西安 710064
4.西安交通大学电子与信息学部，陕西西安 710049

收稿日期:2023-08-26 接受日期:2023-11-03 出版日期:2024-02-29 发布日期:2024-02-29
通讯作者:司利云(1975-)，女，讲师，博士。主要研究方向为电力系统大数据分析、智能算法等。E-mail：siliyun@chd.edu.cn
第一作者:胡欣(1975-)，女，教授，博士。主要研究方向为能源管理、计算机视觉和机器学习等。E-mail：huxin@chd.deu.cn
基金资助:
陕西省秦创原“科学家+工程师”队伍建设项目(2024QGY-KXJ-161);西安市重点产业链项目(23ZDCYJSGG0013-2023);宁夏回族自治区重点研发计划(2022BEG03072)

PCB defect detection method based on fusion of MBAM and YOLOv5

HU Xin¹(), HU Shuai¹, MA Lijun², SI Liyun¹(), XIAO Jian³, YUAN Ye⁴

1. School of Energy and Electrical Engineering, Chang’an University, Xi’an Shaanxi 710064, China
2. State Grid Gansu Electric Power Company TianShui Power Supply Company, Tianshui Gansu 741000, China
3. School of Electronics and Control Engineering, Chang’an University, Xi’an Shaanxi 710064, China
4. Faculty of Electronic and Information Engineering, Xi’an Jiaotong University, Xi’an Shaanxi 710049, China

Received:2023-08-26 Accepted:2023-11-03 Published:2024-02-29 Online:2024-02-29
Contact: SI Liyun (1975-), lecturer, Ph.D. Her main research interests cover power system big data analysis, intelligent algorithms, etc. <break/>E-mail：<email>siliyun@chd.edu.cn</email>
First author：HU Xin (1975-), associate professor, Ph.D. Her main research interests cover energy management, computer vision and machine learning, etc. E-mail：huxin@chd.deu.cn
Supported by:
Shaanxi Province Qin Chuangyuan “Scientist+Engineer” Team Construction Project(2024QGY-KXJ-161);Key Industrial Chain(23ZDCYJSGG0013-2023);Key R&D Projects of Ningxia Hui Autonomous Region(2022BEG03072)

摘要/Abstract

摘要：

随着电子信息产业迅速发展，PCB行业作为电子信息产业的基础，其产品质量对后续生产的电子产品有着决定性影响。针对PCB缺陷目标较小，缺陷类型多，特征不明显，在实际生产过程中易产生误检、漏检等问题，提出了一种多分支注意力MBAM模块方法，在3个不同维度对特征图进行关注，以增强特征提取的能力，对缺陷区域给予更多的注意力表示。通过改进YOLOv5结构，将MBAM与YOLOv5网络结合，有效的提升了对PCB中小目标的检测性能。最后通过在网络不同位置添加MBAM模块进行对比实验，选取了最佳的添加位置。通过在PCB缺陷数据集上的实验结果表明，改进后的PCB缺陷检测算法具有良好的检测性能，优于其他对比算法，最终的AP达到了96.7%，对比标准YOLOv5的94.7%提高了2个百分点，其他项指标均有涨点，在保持检测速度基本不变的情况下，精准地识别PCB缺陷类型。

长安大学胡欣副教授及其学生胡帅等针对PCB缺陷目标较小、缺陷类型多、特征不明显，在生产过程中易产生误检、漏检等问题，提出了一种基于YOLOv5融合多分支注意力模块的方法，在三个维度对特征图进行关注，增强了特征提取能力，对缺陷区域给予了更多的注意力表示，并通过实验选取了最佳的添加位置，提高了算法的检测精度。

关键词: 目标检测, PCB缺陷, 小目标缺陷, YOLOv5, 多分支注意力模块

Abstract:

With the rapid development of the electronic information industry, the printed circuit board (PCB) industry, serving as its foundation, plays a crucial role in determining the quality of electronic products produced subsequently. Addressing the challenges of small defect target in PCBs, numerous defect types, and indistinct features, which often lead to false detection and missed detection in the actual production process, a multi-branch attention multi-branch attention module (MBAM) module method was proposed. This method focused on the feature map in three different dimensions to enhance feature extraction capabilities and allocate more attention to defect areas. By enhancing the YOLOv5 structure and integrating MBAM with YOLOv5 network, the detection performance for small and medium-sized targets in PCBs was effectively improved. Finally, by comparing the MBAM modules at different locations of the network, the best location was selected. The experimental results on the PCB defect dataset demonstrated that the improved PCB defect detection algorithm exhibited superior detection performance compared to other algorithms. The final AP reached 96.7%, a 2 percentage points increase over 94.7% of the standard YOLOv5. Other indicators all showed an upward trend, and the algorithm could accurately identify PCB defect types while maintaining the detection speed.

Key words: target detection, PCB defects, small target defects, YOLOv5, multi-branch attention module

中图分类号:

TP391

胡欣, 胡帅, 马丽军, 司利云, 肖剑, 袁晔. 基于融合MBAM与YOLOv5的PCB缺陷检测方法[J]. 图学学报, 2024, 45(1): 47-55.

HU Xin, HU Shuai, MA Lijun, SI Liyun, XIAO Jian, YUAN Ye. PCB defect detection method based on fusion of MBAM and YOLOv5[J]. Journal of Graphics, 2024, 45(1): 47-55.

图/表 15

图1 常见的PCB缺陷((a)漏孔；(b)鼠咬；(c)开路；(d)短路；(e)杂散；(f)杂铜)

Fig. 1 Common PCB defects ((a) Via hole leakage; (b) Mouse bite; (c) Open circuit; (d) Short circuit; (e) Stray; (f) Pseudo copper)

图2 YOLOv5网络结构

Fig. 2 YOLOv5 network diagram

图3 Focus切片操作

Fig. 3 Focus slicing operations

图4 MBAM结构

Fig. 4 MBAM structure

图5 改进的YOLOv5网络Neck部分结构

Fig. 5 Improved YOLOv5 Network Neck Partial Structure

表1 不同算法结果对比

Table 1 Comparison of results of different algorithms

算法	评价指标
算法	添加MBAM	mAP/%	AP50/%	AP75/%	APS/%	APM/%	FPS/(帧/秒)
YOLOv3	×	33.9	89.8	33.9	34.5	37.1	53.3
YOLOv4	×	46.5	93.4	36.1	45.5	50.0	75.8
RetinaNet	×	46.5	95.2	45.7	46.3	51.4	35.2
Faster-RCNN	×	50.1	95.6	46.3	48.7	53.2	34.1
YOLOv5	×	50.6	94.7	45.9	48.1	54.9	110.4
YOLOv3	√	36.3	91.2	40.7	36.1	40.4	52.1
YOLOv4	√	46.9	94.1	39.7	46.9	50.7	74.9
RetinaNet	√	48.2	96.0	46.6	48.3	51.9	34.6
Faster-RCNN	√	51.6	96.2	47.3	49.9	54.7	32.1
本文算法	√	52.3	96.7	48.1	50.6	56.4	109.1

表2 YOLO系列算法对比

Table 2 Comparison of YOLO Series Algorithms

方法	评价指标
方法	mAP/ %	AP50/ %	AP75/ %	APS/ %	APM/ %	FPS/ (帧/秒)
YOLOX	50.3	94.1	47.0	47.8	53.6	72.6
YOLOv6	51.2	94.9	48.1	48.2	54.2	81.4
YOLOv7	50.9	95.7	47.6	49.6	53.7	97.8
YOLOv8	52.0	96.2	50.2	49.1	55.6	112.6
本文算法	52.3	96.7	48.1	50.6	56.4	109.1

图6 损失函数曲线

Fig. 6 Loss function curve

图7 PR曲线

Fig. 7 PR curve

表3 MBAM模块添加在不同位置的结果

Table 3 Results of MBAM modules added in different locations

方法	位置
方法	MBAM	第一个CSP_3	第二个CSP_3	SPP	AP50/%
YOLOv5	×	×	×	×	94.7
YOLOv5	√	√	×	×	95.1
YOLOv5	√	√	√	×	95.8
本文算法	√	√	√	√	96.7

表4 与其他注意力模块对比

Table 4 Comparison with other attention modules

方法	评价指标
方法	mAP/%	AP50/%	AP75/%	APS/%	APM/%	FPS/(帧/秒)
YOLOv5	50.6	94.7	45.9	48.1	54.9	110.4
YOLOv5+SE	51.0	94.9	46.6	47.4	55.6	83.2
YOLOv5+BAM	51.6	94.9	45.8	48.9	55.5	79.3
YOLOv5+CBAM	52.0	95.3	47.2	49.3	56.0	90.9
本文算法	52.3	96.7	48.1	50.6	56.4	109.1

图8 本文算法在Neck层的特征可视化图((a)浅层特征图；(b)中间层特征图；(c)深层特征图)

Fig. 8 The feature visualization diagram of the algorithm in Neck layer in this paper ((a) Shallow feature map; (b) Middle layer feature map; (c) Deep feature map)

表5 各个类别AP50结果对比/%

Table 5 Comparison of AP50 results by category/%

算法	类别
算法	短路	开路	鼠咬	杂散	杂铜	漏孔
YOLOv5	94.9	97.3	98.0	97.6	97.4	98.9
本文算法	95.7	98.1	98.1	97.7	98.6	98.9

表6 各个类别mAP结果对比/%

Table 6 Comparison of mAP results by category/%

算法	类别
算法	短路	开路	鼠咬	杂散	杂铜	漏孔
YOLOv5	48.8	44.8	49.9	47.5	50.2	55.7
本文算法	50.6	46.5	51.4	48.9	52.6	56.7

图9 不同算法检测结果对比((a) YOLOv3；(b) YOLOv4；(c) YOLOv5；(d)本文算法)

Fig. 9 Comparison of detection results of different algorithms ((a) YOLOv3; (b) YOLOv4; (c) YOLOv5; (d) Ours)

参考文献 25

[1]	吴一全, 赵朗月, 苑玉彬, 等. 基于机器视觉的PCB缺陷检测算法研究现状及展望[J]. 仪器仪表学报, 2022, 43(8): 1-17.
	WU Y Q, ZHAO L Y, YUAN Y B, et al. Research status and the prospect of PCB defect detection algorithm based on machine vision[J]. Chinese Journal of Scientific Instrument, 2022, 43(8): 1-17 (in Chinese).
[2]	卢荣胜, 吴昂, 张腾达, 等. 自动光学(视觉)检测技术及其在缺陷检测中的应用综述[J]. 光学学报, 2018, 38(8): 23-58.
	LU R S, WU A, ZHANG T D, et al. Review on automated optical (visual) inspection and its applications in defect detection[J]. Acta Optica Sinica, 2018, 38(8): 23-58 (in Chinese).
[3]	SANTOYO J, PEDRAZA J C, MEJÍA L F, et al. PCB inspection using image processing and wavelet transform[M]// MICAI 2007:Advances in Artificial Intelligence. Berlin, Heidelberg: Springer Berlin Heidelberg, 2007: 634-639.
[4]	PUTERA S H I, IBRAHIM Z. Printed circuit board defect detection using mathematical morphology and MATLAB image processing tools[C]// 2010 2nd International Conference on Education Technology and Computer. New York: IEEE Press, 2010: V5: 359-V5:363.
[5]	MASHOHOR S, EVANS J R, ARSLAN T. Image registration of printed circuit boards using hybrid genetic algorithm[C]// 2006 IEEE International Conference on Evolutionary Computation. New York: IEEE Press, 2006: 2685-2690.
[6]	MAR N S S, YARLAGADDA P K D V, FOOKES C. Design and development of automatic visual inspection system for PCB manufacturing[J]. Robotics and Computer-Integrated Manufacturing, 2011, 27(5): 949-962. DOI URL
[7]	BAYGIN M, KARAKOSE M, SARIMADEN A, et al. Machine vision based defect detection approach using image processing[C]// 2017 International Artificial Intelligence and Data Processing Symposium. New York: IEEE Press, 2017: 1-5.
[8]	陈世哲. 微电子产品视觉检测中关键技术研究[D]. 哈尔滨: 哈尔滨工业大学, 2006.
	CHEN S Z. Research on the key techniques of microelectronic products vision inspection[D]. Harbin: Harbin Institute of Technology, 2006 (in Chinese).
[9]	李庆, 张丰, 李芬, 等. 一种用于AOI的PCB拼接算法[J]. 华中科技大学学报: 自然科学版, 2011, 39(2): 90-93.
	LI Q, ZHANG F, LI F, et al. A PCB image mosaic algorithm for AOI system[J]. Journal of Huazhong University of Science and Technology: Natural Science Edition, 2011, 39(2): 90-93 (in Chinese).
[10]	宋昀岑. PCB自动光学检测系统基础算法研究[D]. 成都: 电子科技大学, 2014.
	SONG Y C. The study on low-level vision algorithms used in PCB automatic optical inspection systems[D]. Chengdu: University of Electronic Science and Technology of China, 2014 (in Chinese).
[11]	ZHANG F, QIAO N S, LI J F. A PCB photoelectric image edge information detection method[J]. Optik, 2017, 144: 642-646. DOI URL
[12]	GUAN S, GUO F L. A new image enhancement algorithm for PCB defect detection[C]// 2011 International Conference on Intelligence Science and Information Engineering. New York: IEEE Press, 2011: 454-456.
[13]	姚忠伟. 基于机器视觉的PCB缺陷检测算法研究[D]. 哈尔滨: 哈尔滨工业大学, 2013.
	YAO Z W. Automatic optical inspection on printed circuit board[D]. Harbin: Harbin Institute of Technology, 2013 (in Chinese).
[14]	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. DOI URL
[15]	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. DOI URL
[16]	HUANG W B, WEI P. A PCB dataset for defects detection and classification[EB/OL]. [2023-08-01]. https://arxiv.org/abs/1901.08204.pdf.
[17]	唐三立. 基于高分辨率图像在线PCB缺陷检测关键技术研究[D]. 上海: 上海交通大学, 2019.
	TANG S L. Study on the key technique of PCB defect detection based on high-resolution images[D]. Shanghai: Shanghai Jiao Tong University, 2019 (in Chinese).
[18]	杨杰, 张书杰. 基于密集YOLOv3的印刷电路板缺陷识别[J]. 北京邮电大学学报, 2022, 45(5): 42-48.
	YANG J, ZHANG S J. Defect recognition of printed circuit board based on YOLOv3-dense[J]. Journal of Beijing University of Posts and Telecommunications, 2022, 45(5): 42-48 (in Chinese).
[19]	马德鑫. 基于深度学习的PCB缺陷检测方法研究[D]. 长沙: 湖南大学, 2021.
	MA D X. Research on PCB defect detection method based on deep learning[D]. Changsha: Hunan University, 2021 (in Chinese).
[20]	钱万明, 朱红萍, 朱泓知, 等. 基于自适应加权特征融合的PCB裸板缺陷检测研究[J]. 电子测量与仪器学报, 2022, 36(10): 92-99.
	QIAN W M, ZHU H P, ZHU H Z, et al. Research on defect detection of PCB bare board basedon adaptive weighted feature fusion[J]. Journal of Electronic Measurement and Instrumentation, 2022, 36(10): 92-99 (in Chinese).
[21]	颛孙壮志. 基于图像处理的PCB裸板表面缺陷检测系统[D]. 成都: 电子科技大学, 2022.
	ZHUANSUN Z Z. PCB bare board surface defect detection system based on image processing[D]. Chengdu: University of Electronic Science and Technology of China, 2022 (in Chinese).
[22]	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. DOI URL
[23]	何止戈. 基于深度学习方法的PCB图像缺陷检测[D]. 成都: 电子科技大学, 2020.
	HE Z G. Defect of PCB image detection based on deep learning method[D]. Chengdu: University of Electronic Science and Technology of China, 2020 (in Chinese).
[24]	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. [2023-08-01]. https://arxiv.org/abs/2004.10934.pdf.
[25]	CHEN K, WANG J Q, PANG J M, et al. MMDetection: open MMLab detection toolbox and benchmark[EB/OL]. [2023- 08-01]. https://arxiv.org/abs/1906.07155.pdf.

基于融合MBAM与YOLOv5的PCB缺陷检测方法

PCB defect detection method based on fusion of MBAM and YOLOv5

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 25

相关文章 15

编辑推荐

Metrics

本文评价

[1]	李大湘, 吉展, 刘颖, 唐垚. 改进YOLOv7遥感图像目标检测算法[J]. 图学学报, 2024, 45(4): 650-658.
[2]	胡欣, 常娅姝, 秦皓, 肖剑, 程鸿亮. 基于改进YOLOv8和GMM图像点集匹配的双目测距方法[J]. 图学学报, 2024, 45(4): 714-725.
[3]	牛为华, 郭迅. 基于改进YOLOv8的船舰遥感图像旋转目标检测算法[J]. 图学学报, 2024, 45(4): 726-735.
[4]	曾志超, 徐玥, 王景玉, 叶元龙, 黄志开, 王欢. 基于SOE-YOLO轻量化的水面目标检测算法[J]. 图学学报, 2024, 45(4): 736-744.
[5]	宫永超, 沈旭昆. 一种用于互惠目标检测与实例分割的深层架构[J]. 图学学报, 2024, 45(4): 745-759.
[6]	赵磊, 李栋, 房建东, 曹琪. 面向交通标志的改进YOLO目标检测算法[J]. 图学学报, 2024, 45(4): 779-790.
[7]	李跃华, 仲新, 姚章燕, 胡彬. 基于改进YOLOv5s的着装不规范检测算法研究[J]. 图学学报, 2024, 45(3): 433-445.
[8]	张相胜, 杨骁. 基于改进YOLOv7-tiny的橡胶密封圈缺陷检测方法[J]. 图学学报, 2024, 45(3): 446-453.
[9]	魏陈浩, 杨睿, 刘振丙, 蓝如师, 孙希延, 罗笑南. 具有双层路由注意力的YOLOv8道路场景目标检测方法[J]. 图学学报, 2023, 44(6): 1104-1111.
[10]	王大阜, 王静, 石宇凯, 邓志文, 贾志勇. 基于深度迁移学习的图像隐私目标检测研究[J]. 图学学报, 2023, 44(6): 1112-1120.
[11]	李利霞, 王鑫, 王军, 张又元. 基于特征融合与注意力机制的无人机图像小目标检测算法[J]. 图学学报, 2023, 44(4): 658-666.
[12]	郝帅, 赵新生, 马旭, 张旭, 何田, 侯李祥. 基于TR-YOLOv5的输电线路多类缺陷目标检测方法[J]. 图学学报, 2023, 44(4): 667-676.
[13]	邓渭铭, 杨铁军, 李纯纯, 黄琳. 基于神经网络架构搜索的铭牌目标检测方法[J]. 图学学报, 2023, 44(4): 718-727.
[14]	胡欣, 周运强, 肖剑, 杨杰. 基于改进YOLOv5的螺纹钢表面缺陷检测[J]. 图学学报, 2023, 44(3): 427-437.
[15]	毛爱坤, 刘昕明, 陈文壮, 宋绍楼. 改进YOLOv5算法的变电站仪表目标检测方法[J]. 图学学报, 2023, 44(3): 448-455.