TD-YOLO: A Lightweight Detection Algorithm for Tiny Defects in High-Resolution PCBs

被引:4
作者
Ling, Qin [1 ]
Isa, Nor Ashidi Mat [1 ]
机构
[1] Univ Sains Malaysia, Sch Elect & Elect Engn, Engn Campus, Nibong Tebal 14300, Pulau Pinang, Malaysia
关键词
deep learning; lightweight; PCB tiny defect; rapid detection; YOLOv5;
D O I
10.1002/adts.202300971
中图分类号
O [数理科学和化学]; P [天文学、地球科学]; Q [生物科学]; N [自然科学总论];
学科分类号
07 ; 0710 ; 09 ;
摘要
Printing circuit board (PCB) defect inspection precisely and efficiently is an essential and challenging issue. Therefore, based on several improvements upon YOLOv5-nano, a novel lightweight detector named TD-YOLO is proposed to inspect tiny defects in PCBs. First, the lightweight ShuffleNet block is implemented into the backbone to effectively reduce the model weight. Second, novel anchors are designed using modified k-means clustering to accelerate the model convergence and yield superior detection precision. Then, data augmentation strategy is recomposed by rejecting mosaic augmentation to suppress the emergence of extremely tiny targets. Finally, a mighty feature pyramid network namely MPANet, is newly proposed to boost the feature fusion capability of the model. The experiment results denote TD-YOLO achieves the highest 99.5% mean average precision on our dataset, outperforming other state of the arts. Specially, the detection metrics for the smallest two defects, such as spur and mouse bite, are increased by 2.1% and 1.2%, respectively, compared with YOLOv5-nano. Besides, TD-YOLO has only 1.33 million parameters, decreased by 25% than the baseline. Using a mediocre processor, the detection speed is boosted by 20%, reaching 37 frames per second for the input size of 2240x 2240 pixels. A novel lightweight deep learning model named TD-YOLO is proposed based on YOLOv5-nano. TD-YOLO achieves the highest 99.5% mAP@0.5 for tiny PCB defects, outperforming other SOTAs. The detection precision for spur defect is increased by 2.1%. TD-YOLO is lightweight, owning 25% fewer parameters than YOLOv5-nano.The detection speed reaches 37 FPS for high-resolution PCBs, exceeding 20% faster than the original model.image
引用
收藏
页数:16
相关论文
共 53 条
  • [1] Adibhatla V. A., 2020, ELECTRONICS, V9, P9
  • [2] Applying deep learning to defect detection in printed circuit boards via a newest model of you-only-look-once
    Adibhatla, Venkat Anil
    Chih, Huan-Chuang
    Hsu, Chi-Chang
    Cheng, Joseph
    Abbod, Maysam F.
    Shieh, Jiann-Shing
    [J]. MATHEMATICAL BIOSCIENCES AND ENGINEERING, 2021, 18 (04) : 4411 - 4428
  • [3] [Anonymous], 2020, YOLOv5 (2020.05)
  • [4] Bajenescu T.-M., 2021, EEA ELECTROTEH, V69, P53, DOI [10.46904/eea.21.69.2.1108006, DOI 10.46904/EEA.21.69.2.1108006]
  • [5] End-to-end deep learning framework for printed circuit board manufacturing defect classification
    Bhattacharya, Abhiroop
    Cloutier, Sylvain G.
    [J]. SCIENTIFIC REPORTS, 2022, 12 (01)
  • [6] Bochkovskiy A., 2020, YOLOV4 OPTIMAL SPEED, DOI DOI 10.48550/ARXIV.2004.10934
  • [7] Carion N., 2020, P EUR C COMP VIS
  • [8] Chauhan APS, 2011, LECT NOTES ENG COMP, P1597
  • [9] AUTOMATED VISUAL INSPECTION - A SURVEY
    CHIN, RT
    HARLOW, CA
    [J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1982, 4 (06) : 557 - 573
  • [10] Xception: Deep Learning with Depthwise Separable Convolutions
    Chollet, Francois
    [J]. 30TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2017), 2017, : 1800 - 1807