Laser & Optoelectronics Progress, Volume. 62, Issue 17, 1739022(2025)

MGT-Fusion: PCBA Defect Detection Method Based on Texture and Depth Information Fusion (Invited)

Zefang Chen1, Mingyuan Zhong1, Hailong Jing2, Guodong Liu3, Qican Zhang1、*, and Junfei Shen1、**
Author Affiliations
  • 1College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, Sichuan , China
  • 2Sichuan ViSensing Technology Co., Ltd., Chengdu 610065, Sichuan , China
  • 3United Electronics Co., Ltd., Jiangxi, Nanchang 330000, Jiangxi , China
  • show less
    Figures & Tables(9)
    Structure of MGT-Fusion network
    Structure of GFM
    Structure of TFM
    Images of the data acquisition setup. (a) Real picture; (b) 3D structure diagram
    Schematic diagrams of each defect type contained in PCBA dataset
    Qualitative comparison of detection effects by different methods
    • Table 1. Comparison of detection results by different methods on the PCBA defect dataset

      View table

      Table 1. Comparison of detection results by different methods on the PCBA defect dataset

      MethodType of input imageParams /106mAP50 /%F1_Score /%FPS /Hz
      DETRRGB image36.795.34929.41
      Depth image36.790.63899.17
      YOLOv7-lRGB image28.596.059311.28
      Depth image28.591.199011.09
      YOLOv8-mRGB image25.996.829412.13
      Depth image25.992.559112.07
      YOLOv12-mRGB image19.597.129429.42
      Depth image19.592.779129.76
      RT-DETRRGB image32.897.569622.67
      Depth image32.893.059223.64
      MGT-FusionRGB image+depth image43.699.89997.51
    • Table 2. Comparison of APs for different types of defects detection by different methods

      View table

      Table 2. Comparison of APs for different types of defects detection by different methods

      MethodType ofinput imageMissingpartDimension wrongPart tiltPart shift

      Side

      termination

      Upturned

      Insufficient

      solder

      BridgePin offset
      DETRRGB image92.8097.2293.9191.7490.4395.8899.5499.3697.21
      Depth image81.3796.2796.9197.5397.3362.1399.3197.1087.73
      YOLOv7-lRGB image93.3397.6595.6194.4993.8895.0499.0598.5296.89
      Depth image83.4796.2997.2997.4196.9963.8397.7998.3789.33
      YOLOv8-mRGB image96.1997.7598.4297.9393.6495.7698.0396.5697.17
      Depth image79.9893.9797.6597.2998.0779.2198.1199.0089.69
      YOLOv12-mRGB image95.5198.4696.0796.3594.6395.6999.4799.5397.11
      Depth image84.1397.4397.9198.1298.8471.0599.5199.5290.56
      RT-DETRRGB image95.8698.7898.4597.3195.2496.8299.4799.5797.46
      Depth image84.0194.3598.3898.8699.1972.4798.6599.4691.49
      MGT-FusionRGB image+depth image100.00100.0099.8199.9999.63100.0099.9999.6399.94
    • Table 3. Results of ablation experiments

      View table

      Table 3. Results of ablation experiments

      TPTFMGFMParams /106mAP50 /%F1_Score /%
      21.1094.4394
      21.1089.7389
      33.4998.7298
      40.4799.4699
      43.6399.8999
    Tools

    Get Citation

    Copy Citation Text

    Zefang Chen, Mingyuan Zhong, Hailong Jing, Guodong Liu, Qican Zhang, Junfei Shen. MGT-Fusion: PCBA Defect Detection Method Based on Texture and Depth Information Fusion (Invited)[J]. Laser & Optoelectronics Progress, 2025, 62(17): 1739022

    Download Citation

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

    Category: AI for Optics

    Received: May. 4, 2025

    Accepted: Jun. 30, 2025

    Published Online: Sep. 8, 2025

    The Author Email: Qican Zhang (zqc@scu.edu.cn), Junfei Shen (shenjunfei@scu.edu.cn)

    DOI:10.3788/LOP251144

    CSTR:32186.14.LOP251144

    Topics