Optics and Precision Engineering, Volume. 32, Issue 14, 2286(2024)
Solar cell defect detection network combining multiscale feature and attention
Figures & Tables(16)
Table 1. Dataset details
View tableView in ArticleTable 1. Dataset details
类别 Crack Finger Thick_line Star_crack 数量 2 262 2 672 2 173 1 462 示例
Table 2. Ablation experiments of CMFAnet
View tableView in ArticleTable 2. Ablation experiments of CMFAnet
MCD LFAI HFAI P R F mAP Params/M Flops/G FPS — — — 87.79 65.76 74.53 83.75 3.16 8.86 114.5 √ — — 92.62 74.88 82.25 88.37 3.37 8.89 114.4 — √ — 86.67 75.24 80.55 85.42 3.41 8.9 110.5 — — √ 84.88 72.31 78.09 84.79 3.64 8.71 112.7 — √ √ 89.56 82.04 87.52 86.33 3.92 9.27 112.6 √ √ √ 88.12 85.41 86.73 91.40 6.36 11.9 106.3
Table 3. Comparative experiments of different attention
View tableView in ArticleTable 3. Comparative experiments of different attention
方法 P R F APcr APfi APsc APtl mAP Params/M Flops/G FPS v8n-MCD（w/o MCA） 86.84 69.42 76.75 77.98 92.48 88.60 84.14 85.80 3.21 8.86 118.1 v8n-MCD（SE） 90.75 67.90 77.00 75.61 93.68 92.46 84.30 86.51 3.96 8.91 109.8 v8n-MCD（CBAM） 90.05 71.23 79.25 75.55 94.25 92.57 84.44 86.70 4.83 9.04 103.2 v8n-MCD（CA） 92.08 69.51 78.51 76.28 93.94 96.20 81.85 87.07 3.52 8.97 112.6 v8n-MCD（MCA） 92.62 74.88 82.25 78.89 94.56 96.42 84.52 88.37 3.37 8.89 114.4
Table 4. Comparative experiments of different FPN
View tableView in ArticleTable 4. Comparative experiments of different FPN
方法 P R F APcr APfi APsc APtl mAP Params/M Flops/G FPS v8n-PANet 86.24 63.11 72.88 78.48 91.73 84.94 72.17 81.83 3.15 8.86 114.5 v8n-BiFPN 88.57 52.94 66.27 71.28 92.14 78.83 76.43 79.67 3.14 7.41 131.2 v8n-AFPN 90.61 80.84 85.45 79.73 84.91 88.26 85.98 84.72 3.58 8.44 109.6 v8n-CCFM 85.95 73.51 79.24 77.59 89.42 83.47 83.44 83.48 3.21 6.7 127.4 v8n-LFAI&HFAI 89.56 82.04 87.52 76.56 93.79 91.09 83.90 86.33 3.92 9.27 112.6
Table 5. Comparative experiments of different detection networks
View tableView in ArticleTable 5. Comparative experiments of different detection networks
方法 P R F APcr APfi APsc APtl mAP Params/M Flops/G FPS SSD 93.02 34.02 45.75 65.12 74.81 92.66 63.99 74.14 26.29 62.75 49.5 Faster R-cnn 94.47 35.86 48.50 65.29 82.77 93.43 69.73 77.80 28.31 253.5 41.3 EfficientDet 89.74 40.57 53.01 70.46 86.38 94.56 71.29 80.67 15.7 8.4 68.6 YOLOv5s 91.26 61.95 72.51 71.48 92.74 93.25 73.39 82.71 7.28 17.16 62.3 GhostNet 91.55 59.32 71.25 68.69 92.51 83.34 79.13 80.92 5.37 1.52 70.8 YOLOv7 93.86 70.67 79.75 77.41 93.73 94.34 85.23 87.68 37.62 59.53 61.4 YOLOv7-tiny 94.42 38.66 51.25 71.06 85.00 93.43 69.81 79.87 6.07 13.32 82.1 YOLOv8n 87.79 65.76 74.53 77.68 92.63 86.59 78.11 83.75 3.16 8.86 114.5 CMFAnet 88.12 85.41 86.73 87.50 95.90 99.40 83.00 91.40 6.36 11.9 106.3
Tools
Get Citation
Copy Citation Text
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
Paper Information
Category:
Received: Dec. 4, 2023
Accepted: --
Published Online: Sep. 27, 2024
The Author Email: Ying ZHOU (zhouying2007@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20