Optics and Precision Engineering, Volume. 31, Issue 2, 234(2023)
Parallel path and strong attention mechanism for building segmentation in remote sensing images
Figures & Tables(13)
Fig. 6. Comparison of WHU dataset and Massachusetts Buildings dataset
Fig. 8. Comparison of segmentation results in Massachusetts Buildings dataset
Table 1. Comparison of segmentation indicators in WHU dataset
View tableView in ArticleTable 1. Comparison of segmentation indicators in WHU dataset
模型性能对比 MIoU Precision Recall F1-score 参数量/M 计算量/M ResUNet-a 84.08% 92.63% 90.11% 91.35% 43.16 86.36 PSPNet 89.1% 94.5% 93.97% 94.24% 46.72 93.48 ResNet101 88.41% 93.98% 93.71% 93.85% 52.11 104.31 HRNetv2 90.27% 95.23% 94.53% 94.88% 29.54 59.20 SCAttNet 88.1% 94.63% 92.73% 93.67% 29.46 39.65 Ours 90.45% 94.9% 95.07% 94.98% 26.43 52.95
Table 2. Comparison of segmentation indicators in Massachusetts Buildings dataset
View tableView in ArticleTable 2. Comparison of segmentation indicators in Massachusetts Buildings dataset
模型性能对比 MIoU Precision Recall F1-score ResUNet-a 71.38% 86.23% 80.56% 83.3% PSPNet 67.74% 82.26% 79.32% 80.76% ResNet101 67.46% 81.93% 79.25% 80.57% HRNetv2 70.28% 84.69% 80.50% 82.54% SCAttNet 70.36% 85.57% 79.83% 82.61% Ours 72.84% 86.81% 81.91% 84.29%
Table 3. Comparison with existing models on the WHU dataset
View tableView in ArticleTable 3. Comparison with existing models on the WHU dataset
模型性能对比 MIoU Precision Recall F1-score SRI-Net 89.23% 95.67% 93.69% 94.51% DE-Net （90.12 ±0.24）% （95.00 ±0.16）% （94.60 ±0.19）% （94.80 ±0.18）% DS-Net 90.4% 94.85% 95.06% 94.96% AGEDNet 90.29% 94.97% 94.81% 94.9% RBUNet 89.39% 95.56% 93.25% 94.4% Ours 90.45% 94.9% 95.07% 94.98%
Table 4. Comparison of ablation experimental indexes
View tableView in ArticleTable 4. Comparison of ablation experimental indexes
数据集 模型对比 MIOU Precision Recall F1-score 参数量 计算量 WHU baseline 89.49% 94.52% 94.38% 94.45% 23.35 46.78 Baseline+A 90.30% 94.48% 95.33% 94.90% 26.39 52.88 Baseline+P 89.86% 95.76% 93.58% 94.66% 23.37 46.82 PPA-Net 90.45% 94.9% 95.07% 94.98% 26.43 52.95 Massachusetts Buildings baseline 71.52% 86.34% 80.64% 83.39% Baseline+A 72.62% 87.03% 81.43% 84.14% Baseline+P 72.01% 86.48% 81.15% 83.73% PPA-Net 72.84% 86.81% 81.91% 84.29%
Tools
Get Citation
Copy Citation Text
Jianhua YANG, Hao ZHANG, Haiyang HUA. Parallel path and strong attention mechanism for building segmentation in remote sensing images[J]. Optics and Precision Engineering, 2023, 31(2): 234
Paper Information
Category: Information Sciences
Received: Mar. 1, 2022
Accepted: --
Published Online: Feb. 9, 2023
The Author Email: HUA Haiyang (c3i11@sia.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20