• Optics and Precision Engineering
  • Vol. 32, Issue 14, 2286 (2024)
Ying ZHOU1,2,*, Shibo XU1, Haiyong CHEN1,2, and Kun LIU1,2
Author Affiliations
  • 1School of Artificial Intelligence, Hebei University of Technology, Tianjin30030, China
  • 2China Hebei Control Engineering Research Center, Tianjin300130, China
  • show less
    DOI: 10.37188/OPE.20243214.2286 Cite this Article
    Ying ZHOU, Shibo XU, Haiyong CHEN, Kun LIU. Solar cell defect detection network combining multiscale feature and attention[J]. Optics and Precision Engineering, 2024, 32(14): 2286 Copy Citation Text show less
    Architecture of CMFAnet
    Fig. 1. Architecture of CMFAnet
    Architecture diagram of MCA
    Fig. 2. Architecture diagram of MCA
    Architecture of MCD
    Fig. 3. Architecture of MCD
    Architecture of LFAI-Module
    Fig. 4. Architecture of LFAI-Module
    Architecture of HFAI-Module
    Fig. 5. Architecture of HFAI-Module
    confusion matrix
    Fig. 6. confusion matrix
    P-R curve
    Fig. 7. P-R curve
    F-Measure curve
    Fig. 8. F-Measure curve
    Comparison of detection results
    Fig. 9. Comparison of detection results
    Comparison of heat maps
    Fig. 10. Comparison of heat maps
    Comparison of crack detection results
    Fig. 11. Comparison of crack detection results
    类别CrackFingerThick_lineStar_crack
    数量2 2622 6722 1731 462
    示例
    Table 1. Dataset details
    MCDLFAIHFAIPRFmAPParams/MFlops/GFPS
    87.7965.7674.5383.753.168.86114.5
    92.6274.8882.2588.373.378.89114.4
    86.6775.2480.5585.423.418.9110.5
    84.8872.3178.0984.793.648.71112.7
    89.5682.0487.5286.333.929.27112.6
    88.1285.4186.7391.406.3611.9106.3
    Table 2. Ablation experiments of CMFAnet
    方法PRFAPcrAPfiAPscAPtlmAPParams/MFlops/GFPS
    v8n-MCD(w/o MCA)86.8469.4276.7577.9892.4888.6084.1485.803.218.86118.1
    v8n-MCD(SE)90.7567.9077.0075.6193.6892.4684.3086.513.968.91109.8
    v8n-MCD(CBAM)90.0571.2379.2575.5594.2592.5784.4486.704.839.04103.2
    v8n-MCD(CA)92.0869.5178.5176.2893.9496.2081.8587.073.528.97112.6
    v8n-MCD(MCA)92.6274.8882.2578.8994.5696.4284.5288.373.378.89114.4
    Table 3. Comparative experiments of different attention
    方法PRFAPcrAPfiAPscAPtlmAPParams/MFlops/GFPS
    v8n-PANet86.2463.1172.8878.4891.7384.9472.1781.833.158.86114.5
    v8n-BiFPN88.5752.9466.2771.2892.1478.8376.4379.673.147.41131.2
    v8n-AFPN90.6180.8485.4579.7384.9188.2685.9884.723.588.44109.6
    v8n-CCFM85.9573.5179.2477.5989.4283.4783.4483.483.216.7127.4
    v8n-LFAI&HFAI89.5682.0487.5276.5693.7991.0983.9086.333.929.27112.6
    Table 4. Comparative experiments of different FPN
    方法PRFAPcrAPfiAPscAPtlmAPParams/MFlops/GFPS
    SSD93.0234.0245.7565.1274.8192.6663.9974.1426.2962.7549.5
    Faster R-cnn94.4735.8648.5065.2982.7793.4369.7377.8028.31253.541.3
    EfficientDet89.7440.5753.0170.4686.3894.5671.2980.6715.78.468.6
    YOLOv5s91.2661.9572.5171.4892.7493.2573.3982.717.2817.1662.3
    GhostNet91.5559.3271.2568.6992.5183.3479.1380.925.371.5270.8
    YOLOv793.8670.6779.7577.4193.7394.3485.2387.6837.6259.5361.4
    YOLOv7-tiny94.4238.6651.2571.0685.0093.4369.8179.876.0713.3282.1
    YOLOv8n87.7965.7674.5377.6892.6386.5978.1183.753.168.86114.5
    CMFAnet88.1285.4186.7387.5095.9099.4083.0091.406.3611.9106.3
    Table 5. Comparative experiments of different detection networks
    Ying ZHOU, Shibo XU, Haiyong CHEN, Kun LIU. Solar cell defect detection network combining multiscale feature and attention[J]. Optics and Precision Engineering, 2024, 32(14): 2286
    Download Citation