Laser & Optoelectronics Progress, Volume. 62, Issue 8, 0812002(2025)

Defect Detection of PCB Based on Lightweight ADS-YOLOv8n

Qitao Hu and Qijie Zou*
Author Affiliations
  • College of Information Engineering, Dalian University, Dalian 116622, Liaoning , China
  • show less
    Figures & Tables(23)
    Network structure of YOLOv8n
    Network structure of ADS-YOLOv8n
    ADown downsampling structures diagram
    DTFM module structure diagram
    DySample module structure diagram. (a) Sampling based dynamic upsampling; (b) sampling point generator
    SC-bottleneck structure diagram
    SCM structure diagram
    SCConv structure diagram
    MLCA structure diagram
    Types of PCB defects. (a) Missing hole; (b) mouse bite; (c) open circuit; (d) short circuit; (e) spur; (f) spurious copper
    Comparison of recall and mAP@0.5 value curves. (a) Recall; (b) mAP@0.5
    Comparison of class activation maps. (a) Original images; (b) YOLOv8n; (c) ADS-YOLOv8n
    Comparison of defect detection results. (a)‒(f) YOLOv8n defect detection results; (g)‒(l) ADS-YOLOv8n defect detection results
    Comparison of detection results of NEU-DET data set
    • Table 1. Number of PCB defect labels and pictures

      View table

      Table 1. Number of PCB defect labels and pictures

      Defect typeOriginal imageEnhanced image
      Missing hole115460
      Mouse bite115460
      Open circuit116464
      Short circuit116464
      Spur115460
      Spurious copper116464
      Total6932772
    • Table 2. Ablation experiment of ADown module

      View table

      Table 2. Ablation experiment of ADown module

      NetworkRecall /%mAP@0.5 /%Parameter /106GFLOPs /109
      None91.4195.233.08.1
      Backbone92.9396.352.77.6
      Neck92.6296.032.97.9
      Backbone+Neck93.2696.572.67.4
    • Table 3. Ablation experiment

      View table

      Table 3. Ablation experiment

      ModuleRecall /%mAP@0.5 /%Parameter /106GFLOPs /109
      None91.4195.233.08.1
      ADown93.2696.572.67.4
      DTFM92.2696.103.18.6
      SCM92.3596.252.98.0
      WIoUv392.1796.123.08.1
      ADown+DTFM95.6597.212.77.9
      ADown+SCM95.4197.132.47.3
      ADown+DTFM+SCM96.1597.872.57.8
      ADown+DTFM+SCM+WIoUv396.5898.432.57.8
    • Table 4. Comparison of validity experiment results

      View table

      Table 4. Comparison of validity experiment results

      ModuleRecall /%mAP@0.5 /%
      NNI92.0395.88
      DySample92.2696.10
    • Table 5. Horizontal comparison of SCM

      View table

      Table 5. Horizontal comparison of SCM

      ModulemAP@0.5 /%
      SC-bottleneck+CBAM95.92
      SC-bottleneck+ECA96.03
      SC-bottleneck+CA96.12
      SCM96.25
    • Table 6. Comparison of loss function

      View table

      Table 6. Comparison of loss function

      LossmAP@0.5 /%
      CIoU97.87
      DIoU97.33
      EIoU97.46
      SIoU97.75
      WIoUv398.43
    • Table 7. Comparison of AP values of various defects

      View table

      Table 7. Comparison of AP values of various defects

      ModelAP /%
      Missing holeMouse biteOpen circuitShort circuitSpurSpurious copper
      YOLOv8n98.6893.2594.4298.7091.8394.50
      ADS-YOLOv8n99.2597.9398.9199.4495.8399.22
    • Table 8. Comparative experiment of different algorithms

      View table

      Table 8. Comparative experiment of different algorithms

      ModelmAP@0.5 /%Parameter /106GFLOPs /109
      Faster R-CNN87.20136.8
      SSD94.1212.338.8
      FCOS96.2732.1103.3
      Centernet95.3132.770.2
      YOLOv5n94.222.57.1
      YOLOv6n94.344.211.8
      YOLOv7-tiny95.166.013.2
      YOLOv8n95.233.08.1
      YOLOv8s97.3111.128.4
      Reference[1099.147.27.2
      YOLOv5-TGs2898.206.714.4
      Reference[2997.403.79.8
      SimAM-YOLO3098.397.219.2
      YOLO-P3198.805.2
      FFS⁃YOLO3299.226.1
      YOLOv5-HSTE3398.707.848.7
      YOLOv8-PCB3498.372.611.7
      ADS-YOLOv8n98.432.57.8
    • Table 9. Comparison of experimental results on generalization performance

      View table

      Table 9. Comparison of experimental results on generalization performance

      ModelRecall /%mAP@0.5 /%Parameter /106GFLOPs /109
      YOLOv8n71.3270.633.08.1
      ADS-YOLOv8n74.8773.262.57.8
    Tools

    Get Citation

    Copy Citation Text

    Qitao Hu, Qijie Zou. Defect Detection of PCB Based on Lightweight ADS-YOLOv8n[J]. Laser & Optoelectronics Progress, 2025, 62(8): 0812002

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Aug. 28, 2024

    Accepted: Oct. 8, 2024

    Published Online: Apr. 3, 2025

    The Author Email: Qijie Zou (jessie_zou_zou@163.com)

    DOI:10.3788/LOP241923

    CSTR:32186.14.LOP241923

    Topics