Laser & Optoelectronics Progress, Volume. 61, Issue 8, 0812008(2024)

Defect Detection of Photovoltaic Cells Based on Improved YOLOv8

Ying Zhou1,2, Yuze Yan1, Haiyong Chen1,2、*, and Shenghu Pei1
Author Affiliations
  • 1School of Artificial Intelligence and Data Science, Hebei University of Technology, Tianjin 300130, China
  • 2Hebei Control Engineering Technology Research Center, Tianjin 300130, China
  • show less
    Figures & Tables(17)
    Structure of GauGAN
    Images generated by GauGAN. (a) Semantic segmentation mask; (b) images generated when epoch is 100; (c) images generated when epoch is 200
    Structure of YOLOv8
    Structure of CAM
    Structure of MADH
    Structure of two attention mechanisms
    Overall frame
    Partial original dataset
    Comparison of class activation map
    Comparison of predicted results. (a) Large scale crack; (b) star crack; (c) thick line; (d) finger interruption and small scale crack
    • Table 1. Dataset details

      View table

      Table 1. Dataset details

      Defect typeCrackStar crackThick lineFingerTotal
      Small and medium scaleLarge scale

      Before

      expansion

      Label number1028104252118929185491
      Image number90710421977514923497

      After

      expansion

      Label number1821300633118929186861
      Image number170030060077514924867
    • Table 2. Comparison of data enhancement effects

      View table

      Table 2. Comparison of data enhancement effects

      DatasetF1Recall /%Precision /%mAP@0.5 /%Average mAP@0.5 /%Average mAP@0.5∶0.95 /%
      CrackStar crackThick lineFinger
      Before expansion80.2576.4185.1980.4775.8485.1894.0683.8949.50
      After expansion82.7577.2986.9384.0580.6385.8194.3486.2152.00
    • Table 3. CAM improvement experimental results

      View table

      Table 3. CAM improvement experimental results

      ModelParams /106Flops /109F1Recall /%Precision /%mAP@0.5 /%mAP@0.5∶0.95 /%
      YOLOv811.1414.2882.7577.2986.9386.2152.00
      YOLOv8+FAM12.9515.2282.5077.6689.4487.9253.30
      YOLOv8+FBM11.3114.2983.0079.4786.9787.5454.00
      YOLOv8+FBM+FAM13.0415.2382.0078.0988.2588.5653.50
      YOLOv8+FAM+FBM13.0415.2384.5079.5090.5288.7954.00
    • Table 4. MADH improvement experimental results

      View table

      Table 4. MADH improvement experimental results

      ModelParams /106Flops /109F1Recall /%Precision /%mAP@0.5 /%mAP@0.5∶0.95 /%
      YOLOv811.1414.2882.7577.2986.9386.2152.00
      YOLOv8+CA11.0313.9783.7578.3587.4287.6454.70
      YOLOv8+SE10.6913.0483.2580.7686.4785.9952.90
      YOLOv8+CCA10.7113.0784.7582.0687.8787.4452.50
      YOLOv8+CA+CCA10.5912.7684.0081.7986.6088.3554.20
    • Table 5. Ablation experimental results

      View table

      Table 5. Ablation experimental results

      ModelParams /106Flops /109FPS /(frame/s)F1Recall /%Precision /%mAP@0.5 /%mAP@0.5∶0.95 /%
      YOLOv811.1414.2881.2582.7577.2986.9386.2152.00
      YOLOv8+CAM13.0415.2374.0684.5079.5090.5288.7954.00
      YOLOv8+MADH10.5912.7678.9684.0081.7986.6088.3554.20
      YOLOv8-EL13.1315.4373.3084.2579.9090.1289.9054.70
    • Table 6. Comparison test results

      View table

      Table 6. Comparison test results

      ModelParams /106Flops /109F1Recall /%Precision /%mAP@0.5 /%mAP @0.5∶0.95 /%
      Faster R-CNN28.31253.558.8368.8575.0775.8042.30
      YOLO5-s7.0315.8268.7556.6390.5282.8045.40
      YOLO5-m21.0725.2277.5072.0784.2885.4250.60
      YOLOX-s8.9026.8483.0082.1184.7586.4349.40
      YOLOX-m28.5039.4483.7584.3783.4388.0552.80
      YOLOv7-tiny6.0713.3278.7571.2889.2785.8448.30
      YOLOv7-m37.7659.5383.0076.5583.7187.3950.30
      YOLOv8-s11.1414.2882.7577.2986.9386.2152.00
      YOLOv8-m25.8639.4484.6083.8685.5688.7753.70
      YOLOv8-s-EL13.1315.4384.2579.9090.1289.9054.70
      YOLOv8-m-EL29.7141.5186.0087.4989.4190.5855.20
    • Table 7. Generalization performance experimental results

      View table

      Table 7. Generalization performance experimental results

      ModelF1Recall /%Precision /%mAP@0.5 /%mAP@0.5:0.95 /%
      Faster R-CNN78.2580.6882.0179.7051.30
      YOLOv5-s74.5064.3190.0284.1455.90
      YOLOv5-m83.4577.7490.2886.3863.60
      YOLOX-s81.1080.2282.2984.8358.00
      YOLOX-m83.7084.1383.3788.0863.90
      YOLOv7-tiny80.0071.9890.4885.8259.40
      YOLOv7-m87.8086.2189.9390.0268.10
      YOLOv8-s84.7082.7087.2188.0966.00
      YOLOv8-m87.5087.3387.8390.7270.20
      YOLOv8-s-EL84.4580.7488.9990.1167.30
      YOLOv8-m-EL88.1586.4590.1892.8872.30
    Tools

    Get Citation

    Copy Citation Text

    Ying Zhou, Yuze Yan, Haiyong Chen, Shenghu Pei. Defect Detection of Photovoltaic Cells Based on Improved YOLOv8[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0812008

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Jun. 28, 2023

    Accepted: Jul. 31, 2023

    Published Online: Mar. 22, 2024

    The Author Email: Chen Haiyong (haiyong.chen@hebut.edu.cn)

    DOI:10.3788/LOP231622

    Topics