Optics and Precision Engineering, Volume. 32, Issue 2, 286(2024)
Concrete crack segmentation combined with linear guidance and mesh optimization
Figures & Tables(19)
Fig. 1. Linear Guided and Mesh Optimization Joint Network Framework
Fig. 2. Comparison of adaptive single-dimensional pooling and regular pooling
Fig. 6. Crack segmentation results of various models of Crack500 dataset
Fig. 7. Crack segmentation results of various models of Deepcrack537 dataset
Fig. 8. Crack segmentation results of various models of CFD dataset
Table 1. Confusion matrix
Table 2. Experimental results of quantitative analysis of Crack500 dataset
View tableView in ArticleTable 2. Experimental results of quantitative analysis of Crack500 dataset
Methods Metrics Precision Recall F1-score IoU FCN［ 22 ］66.63% 65.03% 65.82% 49.06% DeeplabV3［ 23 ］70.69% 73.72% 72.17% 56.46% SegNet［ 24 ］69.79% 74.45% 72.04% 56.30% PSPNet［ 25 ］73.93% 71.23% 72.55% 56.93% DeepCrack18［ 26 ］66.61% 69.56% 68.05% 51.58% DeepCrack19［ 19 ］70.75% 70.78% 70.76% 54.75% HRNet［ 27 ］69.32% 74.99% 72.04% 56.30% U-Net［ 28 ］69.29% 73.73% 71.44% 55.57% Qu et al［ 16 ］65.40% 69.80% 67.50% / LightCrackNet［ 20 ］75.30% 72.10% 73.70% / FFEDN［ 21 ］71.01% 76.96% 73.87% 58.56% Our model 78.11% 70.64% 74.19% 58.96%
Table 3. Experimental results of quantitative analysis of Deepcrack537 dataset
View tableView in ArticleTable 3. Experimental results of quantitative analysis of Deepcrack537 dataset
Methods Metrics Precision Recall F1-score IoU FCN［ 22 ］74.95% 82.22% 78.42% 64.49% DeeplabV3［ 23 ］82.31% 88.49% 85.29% 74.35% SegNet［ 24 ］83.12% 87.82% 85.41% 74.53% PSPNet［ 25 ］82.67% 88.26% 85.37% 74.48% DeepCrack18［ 26 ］82.21% 89.08% 85.51% 74.69% DeepCrack19［ 19 ］81.10% 88.45% 84.62% 73.34% HRNet［ 27 ］80.70% 86.88% 83.68% 71.94% U-Net［ 28 ］81.33% 88.14% 84.60% 73.31% Qu et al［ 16 ］86.70% 85.00% 85.80% / LightCrackNet［ 20 ］85.50% 87.90% 86.70% / FFEDN［ 21 ］86.73% 85.62% 86.17% 75.70% Our model 84.61% 89.64% 87.05% 77.07%
Table 4. Experimental results of quantitative analysis of CFD dataset
View tableView in ArticleTable 4. Experimental results of quantitative analysis of CFD dataset
Methods Metrics Precision Recall F1-score IoU FCN［ 22 ］54.64% 56.65% 55.63% 38.53% DeeplabV3［ 23 ］60.57% 61.67% 61.11% 44.00% SegNet［ 24 ］61.51% 72.99% 66.76% 50.10% PSPNet［ 25 ］62.67% 74.38% 68.03% 51.54% DeepCrack18［ 26 ］70.27% 69.26% 69.76% 53.56% DeepCrack19［ 19 ］69.95% 70.89% 70.41% 54.34% HRNet［ 27 ］65.32% 70.95% 68.02% 51.53% U-Net［ 28 ］67.98% 73.13% 70.46% 54.39% Our model 71.07% 73.46% 72.24% 56.55%
Table 5. Deepcrack537 dataset ablation experiment results
View tableView in ArticleTable 5. Deepcrack537 dataset ablation experiment results
Methods Metrics Precision Recall F1-score IoU U-Net 81.33% 88.14% 84.60% 73.31% U-Net+MLGM 82.52% 89.00% 85.64% 74.88% U-Net+MDOM 82.54% 89.01% 85.65% 74.91% U-Net+MAM 84.30% 86.35% 85.31% 74.38% Our model 84.61% 89.64% 87.05% 77.07%
Table 6. Crack500 dataset ablation experiment results
View tableView in ArticleTable 6. Crack500 dataset ablation experiment results
Methods Metrics Precision Recall F1-score IoU U-Net 69.29% 73.73% 71.44% 55.57% U-Net+MLGM 77.09% 71.14% 74.00% 58.73% U-Net+MDOM 73.18% 73.32% 73.25% 57.79% U-Net+MAM 75.04% 71.91% 73.44% 58.03% Our model 78.11% 70.64% 74.19% 58.96%
Table 7. CFD dataset ablation experiment results
View tableView in ArticleTable 7. CFD dataset ablation experiment results
Methods Metrics Precision Recall F1-score IoU U-Net 67.98% 73.13% 70.46% 54.39% U-Net+MLGM 70.22% 71.17% 70.69% 54.67% U-Net+MDOM 66.29% 75.54% 70.61% 54.57% U-Net+MAM 66.98% 77.42% 71.82% 56.03% Our model 71.07% 73.46% 72.24% 56.55%
Table 8. Comparative experimental results of Deepcrack537 dataset
View tableView in ArticleTable 8. Comparative experimental results of Deepcrack537 dataset
Metrics Precision Recall F1-score IoU 1，0.5，0.25 84.61% 89.64% 87.05% 77.07% 1，0.75，0.5，0.25 81.36% 88.93% 84.98% 73.88%
Table 9. Comparative experimental results of CFD dataset
View tableView in ArticleTable 9. Comparative experimental results of CFD dataset
Metrics Precision Recall F1-score IoU 1，0.5，0.25 71.07% 73.46% 72.24% 56.55% 1，0.75，0.5，0.25 65.78% 74.57% 69.90% 53.73%
Table 10. Number of parameters and number of floating-point operations
View tableView in ArticleTable 10. Number of parameters and number of floating-point operations
Methods FLOPs/G Params/M FCN［ 22 ］80.51 15.31 DeeplabV3［ 23 ］161.50 39.04 SegNet［ 24 ］160.56 29.44 PSPNet［ 25 ］201.30 53.32 DeepCrack18［ 26 ］548.01 30.91 DeepCrack19［ 19 ］80.46 14.72 HRNet［ 27 ］90.74 29.53 U-Net［ 28 ］160.76 17.26 Our model 170.49 28.24
Tools
Get Citation
Copy Citation Text
Guanghui LIU, Jian CHEN, Yuebo MENG, Shengjun XU. Concrete crack segmentation combined with linear guidance and mesh optimization[J]. Optics and Precision Engineering, 2024, 32(2): 286
Paper Information
Category:
Received: Jun. 2, 2023
Accepted: --
Published Online: Apr. 2, 2024
The Author Email: LIU Guanghui (guanghuil@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20