随着电子产品高度集成化，印刷电路板（Printed Circuit Board,PCB）的生产也更加 细化、走线结构更加复杂。在实际的工业生产过程中，任何一个环节出现问题或者瑕疵， 都会导致 PCB 由于缺陷问题而不能流入市场。传统的缺陷检测方法如人工目测与功能性 测试等已经难以支撑在工业流水线上对 PCB 进行高效准确地检测。基于上述背景，PCB 表面缺陷的自动化检测已经成为了制造业发展的方向之一，目前主流的缺陷检测手段是 自动光学检测（Automated Optical Inspection,AOI）技术，其主要优点在于具有较高的检 测精度与检测效率。AOI 技术的核心内容是图像处理与算法分析模块，因此，本文将对 AOI 技术的核心内容展开深入讨论与研究，其主要研究内容如下： （1）图像采集系统的搭建与图像预处理算法的选取：主要通过工业相机、镜头、光 源以及载物台等硬件设备完成图像采集系统的搭建工作，其次针对图像采集系统获取的 PCB 图像质量不高的问题，分别选取直方图均衡化、白平衡算法等图像预处理方法进行 实验讨论与分析，通过主观视觉评价与客观视觉评价的指标结果反馈，最终选取白平衡 算法用于 PCB 图像的预处理。 （2）图像拼接算法的分析与优化：针对图像采集系统单次无法获取一张完整的高分 辨 PCB 图像，提出使用图像拼接技术解决上述问题，主要以尺度不变特征变换（Scaleinvariant Feature Transform,SIFT）算法为主进行分析与讨论，提出使用分块策略对 SIFT 算法进行相关改进，从实验的反馈结果可以看出，本文所提出的改进拼接算法在拼接准 确度与拼接效率上均得到了一定的改善，同时还具有一定的鲁棒性。 （3）PCB 缺陷检测算法的研究与改进：对基于深度学习的 PCB 缺陷检测方法进行 归纳总结，主要包含 SSD、YOLO 系列、R-CNN、Fast R-CNN、Faster R-CNN、Transformer 等相关模型；其次构建 PCB 缺陷检测数据集，通过优化聚类方法、更换主干特征提取网 络、添加注意力机制与卷积模块等有效手段，提出了一种性能较好的 Transformer-YOLO 模型，其检测准确率可达 97.04％，优于目前大部分主流的缺陷检测模型。

With the high integration of electronic products, the production of printed circuit boards (PCBs) has become more refined and the wiring structure has become more complex. In the actual industrial production process, any problem or defect in any link will lead to PCB being unable to enter the market due to defect issues. Traditional defect detection methods such as manual visual inspection and functional testing are no longer able to support efficient and accurate detection of PCBs on industrial assembly lines. Based on the above background, automated detection of PCB surface defects has become one of the directions of manufacturing development. Currently, the mainstream defect detection method is Automated Optical Inspection (AOI) technology, which has the main advantage of high detection accuracy and efficiency. The core content of AOI technology is the image processing and algorithm analysis module. Therefore, this article will conduct in-depth discussion and research on the core content of AOI technology, and its main research content is as follows: (1) Construction of image acquisition system and selection of image pre-processing algorithm: the construction of image acquisition system is mainly completed by industrial camera, lens, light source, stage and other hardware equipment. Secondly, in view of the low quality of PCB images obtained by the image acquisition system, image pre-processing methods such as histogram equalization and white balance algorithm are selected for experimental discussion and analysis, Through the feedback of subjective and objective visual evaluation indicators, the white balance algorithm was ultimately selected for PCB image preprocessing. (2) Analysis and optimization of image stitching algorithm: In response to the inability of the image acquisition system to obtain a complete high-resolution PCB image at a single time, image stitching technology is proposed to solve the above problem. The analysis and discussion mainly focus on the Scale Invariant Feature Transform (SIFT) algorithm, and a blocking strategy is proposed to improve the SIFT algorithm. From the feedback results of the experiment, it can be seen that:, The improved stitching algorithm proposed in this article has achieved certain improvements in stitching accuracy and efficiency, while also possessing a certain degree of robustness. (3) Research and improvement of PCB defect detection algorithms: Summarize and summarize PCB defect detection methods based on deep learning, mainly including SSD, YOLO series, R-CNN, Fast R-CNN, Fast R-CNN, Transformer and other related models; Secondly, a PCB defect detection dataset was constructed. By optimizing clustering methods, replacing the backbone feature extraction network, and adding attention mechanisms and convolutional modules, a high-performance Transformer YOLO model was proposed, with a detection accuracy of 97.04%, which is superior to most mainstream defect detection models at present.

[1]Angelopoulos A, Michailidis E T, Nomikos N, et al. Tackling faults in the industry 4.0 era—a survey of machine-learning solutions and key aspects[J]. Sensors, 2019, 20(1): 109-142.

[2]Powell D, Magnanini M C, Colledani M, et al. Advancing zero defect manufacturing: A state-of-the-art perspective and future research directions[J]. Computers in Industry, 2022, 136: 103596-103606.

[3]Psarommatis F, May G, Dreyfus P A, et al. Zero defect manufacturing: state-of-the-art review, shortcomings and future directions in research[J]. International journal of production research, 2020, 58(1): 1-17.

[4]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(08):1-17.

[5]Michalska M. Overview of AOI use in surface-mount technology control[J]. Informatyka, Automatyka, Pomiary w Gospodarce i Ochronie Środowiska, 2020, 10(4): 61-64.

[6]朱云,凌志刚,张雨强.机器视觉技术研究进展及展望[J].图学学报,2020,41(06):871-890.

[7]鹿亚琴. HCI-A3型PCB板缺陷检测系统设计与实现[D].电子科技大学,2019.

[8]Li M, Yao N, Liu S, et al. Multisensor image fusion for automated detection of defects in printed circuit boards[J]. IEEE Sensors Journal, 2021, 21(20): 23390-23399.

[9]Li C Y, Guo J C, Cong R M, et al. Underwater image enhancement by dehazing with minimum information loss and histogram distribution prior[J]. IEEE Transactions on Image Processing, 2016, 25(12): 5664-5677.

[10]Li G, Yang Y, Qu X, et al. A deep learning based image enhancement approach for autonomous driving at night[J]. Knowledge-Based Systems, 2021, 213: 106617-106630.

[11]Rahman T, Khandakar A, Qiblawey Y, et al. Exploring the effect of image enhancement techniques on COVID-19 detection using chest X-ray images[J]. Computers in biology and medicine, 2021, 132: 104319-104334.

[12]王利娟,常霞,任旺.基于加权直方图均衡化彩色图像增强仿真[J].计算机仿真,2021,38(12):126-131.

[13]杨嘉能,李华,田宸玮等.基于自适应校正的动态直方图均衡算法[J].计算机工程与设计,2021,42(05):1264-1270.

[14]康利娟,陈先桥.基于多级直方图形状分割的图像对比度增强技术[J].计算机应用与软件,2022,39(03):207-212.

[15]Jebadass J R, Balasubramaniam P. Low contrast enhancement technique for color images using interval-valued intuitionistic fuzzy sets with contrast limited adaptive histogram equalization[J]. Soft Computing, 2022, 26(10): 4949-4960.

[16]Lin Y H, Lu Y C. Low-light enhancement using a plug-and-play Retinex model with shrinkage mapping for illumination estimation[J]. IEEE Transactions on Image Processing, 2022, 31: 4897-4908.

[17]Li C, Anwar S, Hou J, et al. Underwater image enhancement via medium transmission-guided multi-color space embedding[J]. IEEE Transactions on Image Processing, 2021, 30: 4985-5000.

[18]Wang F, Zhang B, Zhang C, et al. Low-light image joint enhancement optimization algorithm based on frame accumulation and multi-scale Retinex[J]. Ad Hoc Networks, 2021, 113: 102398-102404.

[19]Ma J, Jiang X, Fan A, et al. Image matching from handcrafted to deep features: A survey[J]. International Journal of Computer Vision, 2021, 129: 23-79.

[20]Lowe D G. Distinctive image features from scale-invariant keypoints[J]. International journal of computer vision, 2004, 60: 91-110.

[21]Bay H, Ess A, Tuytelaars T, et al. Speeded-up robust features (SURF)[J]. Computer vision and image understanding, 2008, 110(3): 346-359.

[22]Ye F, Hong Z, Lai Y, et al. Multipurification of matching pairs based on ORB feature and PCB alignment case study[J]. Journal of Electronic Imaging, 2018, 27(3): 33029-33043.

[23]卫保国,张玉兰,周佳明.图像匹配中的特征点筛选方法[J].计算机工程与应用,2021,57(03):208-214.

[24]胡涛,茅健.基于降采样和改进Shi-Tomasi角点检测算法的PCB图像拼接[J].电子测量技术,2021,44(22):134-140.

[25]张学峰,马行,穆春阳.基于四叉树的SIFT与K-D树融合的图像匹配研究[J].电子测量技术,2021,44(22):121-127.

[26]陈宁,刘志坚,苏雪平,等.基于改进的SIFT算法的集成电路图像拼接[J].国外电子测量技术,2021,40(06):159-164.

[27]唐颖复,王忠静,张子雄.基于改进SIFT和SURF算法的沙丘图像配准[J].清华大学学报(自然科学版),2021,61(02):161-169.

[28]刘通,胡亮,王永军,等.基于卷积神经网络的卫星遥感图像拼接[J].吉林大学学报(理学版),2022,60(01):99-108.

[29]王少杰,武文波,徐其志.VGG与DoG结合的光学遥感影像精确配准方法[J].航天返回与遥感,2021,42(05):76-84.

[30]Li J, Gu J, Huang Z, et al. Application research of improved YOLO V3 algorithm in PCB electronic component detection[J]. Applied Sciences, 2019, 9(18): 3750-3770.

[31]Liu G, Wen H. Printed circuit board defect detection based on MobileNet-Yolo-Fast[J]. Journal of Electronic Imaging, 2021, 30(4): 43004-43022.

[32]Hu B, Wang J. Detection of PCB surface defects with improved faster-RCNN and feature pyramid network[J]. Ieee Access, 2020, 8: 108335-108345.

[33]Zeng N, Wu P, Wang Z, et al. A small-sized object detection oriented multi-scale feature fusion approach with application to defect detection[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-14.

[34]文斌,胡晖,杨超.基于LWN-Net的PCB缺陷检测算法[J/OL]．计算机集成制造系统，2023-04-14.

[35]苏佳,贾欣雨,侯卫民.基于YOLO-J的PCB缺陷检测算法[J/OL]．计算机集成制造系统，2023-04-14.

[36]汤伟,王锦韫,冯波.基于机器视觉的纸病诊断系统光源设计方案研究[J].中国造纸,2022,41(10):67-71.

[37]Huang T, Dong W, Xie X, et al. Mixed noise removal via Laplacian scale mixture modeling and nonlocal low-rank approximation[J]. IEEE Transactions on Image Processing, 2017, 26(7): 3171-3186.

[38]李凌杰,陈菲菲.基于改进直方图的红外图像增强方法[J].航空兵器,2022,29(02):101-105.

[39]冯迎宾,刘文竹,于伟经等.变压器油下图像增强算法[J].控制与决策,2023,38(04):1101-1108.

[40]孔二伟,张亚邦,李佳悦等.面向煤矿井下低光照图像的增强方法[J].工矿自动化,2023,49(04):62-70.

[41]王林,王超凡.差分信息熵在拼接图像质量评估中的应用[J].计算机仿真,2020,37(04):265-269.

[42]樊彦国,柴江龙,许明明,等.基于ORB与RANSAC融合改进的图像配准[J].光学精密工程,2019,27(03):702-717.

[43]姚伟,胡虹.基于SIFT的自适应旋转图像无缝拼接算法[J].南京理工大学学报,2019,43(02):250-254.

[44]王朝卿,沈小林,李磊.图像相似度计算算法分析[J].现代电子技术,2019,42(09):31-35.

[45]周昌,李向利,李俏霖等.基于余弦相似度的稀疏非负矩阵分解算法[J].计算机科学,2020,47(10):108-113.

[46]Liu W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector [C]//European Conference on Computer Vision(ECCV),Amsterdam,Netherlands:Springer,2016:21-37.

[47]Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition(CVPR). Las Vegas, NV, USA, 2016: 779-788.

[48]Redmon J, Farhadi A. YOLO9000: Better, faster, stronger[C]//Computer Vision and Pattern Recognition(CVPR). 2017:7263-7271.

[49]Redmon J, Farhadi A. YOLOv3: An Incremental Improvement[C]//Proceedings of the IEEE conference on computer vision and pattern recognition(CVPR). 2018: 79-86.

[50]Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4:Optimal Speed and Accuracy of Object Detection[J]. arXiv preprint arXiv,2020,2004:10934-10950.

[51]GIRSHICK R. Fast R-CNN[C]//Proceedings of 2015 IEEE International Conference on Computer Vision(ICCV). Piscataway,NJ,USA:IEEE,2015,1440-1448.

[52]REN S, HE K,GIRSCHICK R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.

[53]Chien-Yao Wang, Hong-Yuan Mark Liao, Yueh-Hua Wu, et al. CSPNet: A new backbone that can enhance learning capability of cnn[C]//CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW),Seattle,USA:IEEE,2020:1571-1580.

[54]Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16x16 words: Transformers for image recognition at scale[J]. arXiv preprint arXiv,2020,2010:11929-11939.