Opto-Electronic Engineering, Volume. 51, Issue 11, 240220-1(2024)

A solar cell defect detection model optimized and improved based on YOLOv8

Ziran Peng1...2,*, Siyuan Wang1,2, and Shenping Xiao12 |Show fewer author(s)
Author Affiliations
  • 1School of Electrical and Information Engineering, Hunan University of Technology, Zhuzhou, Hunan 412007, China
  • 2Hunan Key Laboratory of Electric Drive Control and Intelligent Equipment, Zhuzhou, Hunan 412007, China
  • show less
    Figures & Tables(18)
    YOLOv8n network structure
    Structure of SCI
    Specific algorithm implementation flow
    Schematic diagram of SPD with scale=2
    Conv modules in different positions of the Backbone part
    Schematic diagram of four features of fusion structure
    Calculation of MPDIOU loss function
    Improved EL-YOLO model
    Diagram of three major defect types
    Visual comparison of different enhancement methods
    Comparison of detection effect between the proposed algorithm and YOLOv8n
    Comparison of P, R, and PR curves between the algorithm in this paper and YOLOv8n
    Comparison of loss functions
    Comparison of thermal map effects
    • Table 1. Comparative experimental results of adding attention mechanisms at multiple locations

      View table
      View in Article

      Table 1. Comparative experimental results of adding attention mechanisms at multiple locations

      替换位置权重/MB参数量/(106)GFlopsmAP/%
      YOLOv8n6.33.0068.194.7
      2,3,4,55.72.8187.495.1
      2,55.82.8907.895.3
      55.82.8948.095
      46.23.0668.094.5
      36.33.1098.094.7
      2 (本文)6.23.0027.996.0
    • Table 2. Enhancement algorithm comparison experiment

      View table
      View in Article

      Table 2. Enhancement algorithm comparison experiment

      算法准确率/%召回率/%平均精度/%
      RAW93.191.091.0
      AGT93.590.491.8
      ZERO-DCE95.690.793.2
      SCI (本文)96.591.194.7
    • Table 3. Results of the ablation experiment

      View table
      View in Article

      Table 3. Results of the ablation experiment

      算法模型SSBM权重/MB参数量/106计算量/GFlopsFPSmAP@(0.5)/%
      YOLOv8n6.33.0068.115394.7
      YOLOv8n-S6.23.0027.915596.0
      YOLOv8n-SB6.33.0068.114795.4
      YOLOv8n-M6.33.0068.115495.5
      YOLOv8n-S+SB6.23.0027.914896.2
      YOLOv8n-S+M6.23.0027.915896.4
      YOLOv8n-SB+M6.33.0068.115096.0
      EL-YOLO (本文)6.23.0027.915596.9
    • Table 4. Comparison of results of multiple detection algorithms

      View table
      View in Article

      Table 4. Comparison of results of multiple detection algorithms

      算法模型权重/MB参数量/106计算量/GFlopsFPSmAP@(0.5)/%
      YOLOv5s14.57.01815.811889.8
      YOLOv774.837.194105.17184.6
      YOLOv8n6.33.0068.115394.7
      YOLOv8s22.511.13728.810495.3
      YOLOv8m52.025.90279.37393.6
      YOLOv10s31.47.221.611996.2
      RT-DETR63.132.8108.24696.0
      Gold-YOLO43.121.546.08296.4
      Faster-R-CNN112.741.32524.2894.6
      SSD99.135.87312.74481.3
      EL-YOLO(本文)6.23.0197.915596.9
    Tools

    Get Citation

    Copy Citation Text

    Ziran Peng, Siyuan Wang, Shenping Xiao. A solar cell defect detection model optimized and improved based on YOLOv8[J]. Opto-Electronic Engineering, 2024, 51(11): 240220-1

    Download Citation

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

    Category: Article

    Received: Sep. 16, 2024

    Accepted: Nov. 4, 2024

    Published Online: Jan. 24, 2025

    The Author Email: Peng Ziran (彭自然)

    DOI:10.12086/oee.2024.240220

    Topics