Optics and Precision Engineering, Volume. 32, Issue 14, 2256(2024)

Visual inspection of soldering defects on board surfaces against complex backgrounds

Liying ZHU... Sen WANG*, Aiping SHEN and Xuangang LI |Show fewer author(s)
Author Affiliations
  • Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming650500, China
  • show less
    Figures & Tables(21)
    Architecture of PCBNet
    Dilation and extrusion convolution
    DeConv receptive field
    Architecture of DeCSPlayer
    SPD-Conv
    Subtle feature enhancement module
    Schematic diagram of feature extraction in the PE module
    Schematic diagram of PCB surface soldering defects
    Image acquisition and detection system
    Visualization of detection results for each algorithm
    Visualization of detection results for each algorithm
    • Table 1. Dilatation rate configuration table in the backbone

      View table
      View in Article

      Table 1. Dilatation rate configuration table in the backbone

      NumberDilationLayersOutput size
      0-Focus-320×320×32
      1-SPD-Conv-160×160×64
      21DeCSPlayer-160×160×64
      3-SPD-Conv280×80×128
      42DeCSPlayer-80×80×128
      5-SPD-Conv240×40×256
      63DeCSPlayer-40×40×256
      7-SPD-Conv220×20×512
      8-SPP-20×20×512
      95DeCSPlayer-20×20×512
    • Table 2. Camera parameters during data collection

      View table
      View in Article

      Table 2. Camera parameters during data collection

      项目参数
      相机型号MV-CS060-10UC-PRO
      光圈F4
      焦距8 mm
      光源类型Ring light
      曝光时间50 000 μs
      分辨率3 072×2 048
      物距100 mm
      视野50 mm×35 mm
    • Table 3. Number of defects in various categories in the dataset

      View table
      View in Article

      Table 3. Number of defects in various categories in the dataset

      缺陷类别缺陷数目
      Aiguille2 487
      Dissymmetry1 667
      Holes1 923
      Interconnect pad4 069
    • Table 4. Results of the ablation study

      View table
      View in Article

      Table 4. Results of the ablation study

      方法图像尺寸mAP0.5/%mAP0.5:0.95/%Params/MbFPS
      YOLOX-S64092.560.88.94106
      YOLOX-S+SFEM64093.664.79.7389
      YOLOX-S+DeConv64094.564.69.3690
      YOLOX-S+SPD-Conv64094.364.913.64100
      YOLOX-S+SPD-Conv+SFEM64094.864.714.4390
      YOLOX-S+SPD-Conv+SFEM+DeConv64095.365.014.8586
    • Table 5. Impact of different r values in DeConv on the optimal model accuracy

      View table
      View in Article

      Table 5. Impact of different r values in DeConv on the optimal model accuracy

      方法mAP0.5/%mAP0.5:0.95/%Params /MbFPS
      r=294.664.715.9285
      r=395.365.014.8586
      r=495.164.914.5487
      r=594.765.114.4187
    • Table 6. Impact of different dilation values in the backbone of PCBNet on the optimal model accuracy

      View table
      View in Article

      Table 6. Impact of different dilation values in the backbone of PCBNet on the optimal model accuracy

      空洞率

      mAP0.5

      /%

      mAP0.5:0.95

      /%

      Params

      /Mb

      FPS
      2,2,2,295.065.414.8587
      1,2,1,295.265.314.8588
      1,6,12,1895.365.014.8586
      1,2,5,794.965.014.8586
      1,2,3,595.465.514.8583
      1,2,5,194.965.014.8587
      1,2,9,1295.165.214.8586
    • Table 7. Performance statistics of the model with different numbers of training samples

      View table
      View in Article

      Table 7. Performance statistics of the model with different numbers of training samples

      样本数

      mAP0.5

      /%

      mAP0.5:0.95

      /%

      Params

      /Mb

      FPS
      80%95.465.514.8587
      70%95.265.314.8587
      60%95.065.214.8587
    • Table 8. Results compared to other methods(For statistical convenience, aiguille is simplified as A, disymmetry is simplified as D, holes is simplified as H, and Interconnect pad is simplified as I-P)

      View table
      View in Article

      Table 8. Results compared to other methods(For statistical convenience, aiguille is simplified as A, disymmetry is simplified as D, holes is simplified as H, and Interconnect pad is simplified as I-P)

      模型主干AP0.5(%)

      mAP0.5

      /%

      mAP0.75

      /%

      mAP0.5∶0.95

      /%

      FPS

      Params

      /Mb

      ADHI-P
      Faster-RCNNresnet5061.863.866.785.369.415.328.530136.75
      CenterNetresnet5088.791.795.392.792.161.055.210132.67
      RetinaNetresnet5083.887.991.892.989.161.054.95336.39
      EfficientDetv1EfficientNet-B050.868.871.086.269.225.432.7333.83
      YOLOV4-Tiny63.784.889.689.181.828.337.91925.88
      YOLOV5-sC380.880.992.091.686.356.051.0927.07
      YOLOV8-sC2f90.993.994.994.493.576.863.411411.14
      CFPNet-87.790.696.196.592.771.960.88913.14
      YOLOX-s-86.792.195.695.592.572.360.81068.94
      YOLOX-m-90.892.197.396.493.875.563.07425.28
      YOLO-PCB-54.080.089.564.271.965.852.6958.9
      MDVI-75.887.794.093.987.843.346.016112.22
      RTDD-LCD-68.886.893.992.885.634.941.67325.08
      PCBNet-91.294.897.498.295.479.065.58314.85
    • Table 9. Dataset splitting

      View table
      View in Article

      Table 9. Dataset splitting

      任务图像尺寸数目
      训练600×600×38 460
      验证600×600×3961
      测试600×600×31 067
    • Table 10. Results compared to other methods

      View table
      View in Article

      Table 10. Results compared to other methods

      模型模型版本主干mAP0.5/%mAP0.75/%mAP0.5∶0.95/%FPSParams/Mb
      Faster-RCNN-resnet5072.611.228.430136.75
      CenterNet-resnet5097.746.352.110132.67
      RetinaNet-resnet5092.743.449.35336.39
      EfficientDetv1-EfficientNet-B085.638.643.4333.83
      YOLOV4-Tiny-97.232.646.61925.88
      YOLOV5sC398.360.056.6927.07
      YOLOV8sC2f99.278.465.211411.14
      CFPNets-98.865.359.38913.14
      YOLOxs-98.259.657.01068.94
      YOLOXm-99.186.776.57425.28
      YOLO-PCB--96.771.561.3958.9
      MDVI--98.241.450.516112.22
      RTDD-LCD--97.639.049.17325.08
      PCBNet--99.588.773.28314.85
    Tools

    Get Citation

    Copy Citation Text

    Liying ZHU, Sen WANG, Aiping SHEN, Xuangang LI. Visual inspection of soldering defects on board surfaces against complex backgrounds[J]. Optics and Precision Engineering, 2024, 32(14): 2256

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category:

    Received: Mar. 13, 2024

    Accepted: --

    Published Online: Sep. 27, 2024

    The Author Email: WANG Sen (wangsen0401@126.com)

    DOI:10.37188/OPE.20243214.2256

    Topics