Laser & Optoelectronics Progress, Volume. 61, Issue 4, 0428004(2024)
Building Extraction from Remote Sensing Image Based on Multi-Module
Figures & Tables(15)
Fig. 3. Module improvements. (a) Atrous spatial pyramid pooling module; (b) multi-scale feature enhancement module
Fig. 6. Comparison results of Massachusetts Building dataset. (a) Images; (b) labels; (c) FCN; (d) SegNet; (e) Unet; (f) Unet++; (g) MM-Unet
Fig. 7. Comparison results of WHU Building dataset. (a) Images; (b) labels; (c) FCN; (d) SegNet; (e) Unet; (f) Unet++; (g) MM-Unet
Fig. 8. Comparison results of ISPRS Vaihingen dataset. (a) Images; (b) labels; (c) FCN; (d) SegNet; (e) Unet; (f) Unet++; (g) MM-Unet
Table 1. Experimental dataset
View tableTable 1. Experimental dataset
Dataset Resolution /m Train number Validation number Test number Massachusetts Building 1.00 7416 288 90 WHU Building 0.30 4736 1036 2416 ISPRS Vaihingen 0.09 2752 736 306
Table 2. Accuracy comparison of Massachusetts Building dataset
View tableTable 2. Accuracy comparison of Massachusetts Building dataset
Model ROA Rprecision Rrecall sF1 RIoU FCN 92.78 80.53 80.70 80.61 67.52 SegNet 93.00 81.83 81.68 81.75 69.87 Unet 93.76 83.47 81.90 82.68 71.21 Unet++ 94.14 85.52 82.47 83.97 72.37 MM-Unet 94.46 86.39 83.12 84.72 73.42
Table 3. Accuracy comparison of WHU Building dataset
View tableTable 3. Accuracy comparison of WHU Building dataset
Model ROA Rprecision Rrecall sF1 RIoU FCN 98.64 93.65 94.20 93.92 88.54 SegNet 98.62 93.90 94.00 93.95 88.59 Unet 98.68 93.97 94.23 94.10 88.86 Unet++ 98.79 94.90 93.91 94.40 89.39 MM-Unet 98.85 95.35 94.25 94.80 90.11
Table 4. Accuracy comparison of ISPRS Vaihingen dataset
View tableTable 4. Accuracy comparison of ISPRS Vaihingen dataset
Model ROA Rprecision Rrecall sF1 RIoU FCN 95.80 92.56 90.33 91.43 84.57 SegNet 95.46 91.98 90.09 91.02 83.53 Unet 95.57 91.87 90.26 91.05 83.66 Unet++ 95.63 91.18 91.64 91.40 84.17 MM-Unet 95.95 92.95 91.68 92.31 85.21
Table 5. Ablation experiments of Massachusetts Building dataset
View tableTable 5. Ablation experiments of Massachusetts Building dataset
Model ROA Rprecision Rrecall sF1 RIoU Unet 93.76 83.47 81.90 82.68 71.21 Unet+DAM 94.08 85.46 82.05 83.72 72.16 Unet+DAM+MFCM 94.23 86.15 82.97 84.53 73.15 MM-Unet 94.46 86.39 83.12 84.72 73.42
Table 6. Ablation experiments of WHU Building dataset
View tableTable 6. Ablation experiments of WHU Building dataset
Model ROA Rprecision Rrecall sF1 RIoU Unet 98.68 93.97 94.23 94.10 88.86 Unet+DAM 98.73 94.55 94.09 94.32 89.25 Unet+DAM+MFCM 98.80 95.09 94.05 94.57 89.70 MM-Unet 98.85 95.35 94.25 94.80 90.11
Table 7. Ablation experiments of ISPRS Vaihingen dataset
View tableTable 7. Ablation experiments of ISPRS Vaihingen dataset
Model ROA Rprecision Rrecall sF1 RIoU Unet 95.57 91.87 90.26 91.05 83.66 Unet+DAM 95.84 92.06 91.25 91.65 84.59 Unet+DAM+MFCM 95.90 92.74 90.93 91.82 84.88 MM-Unet 95.95 92.95 91.68 92.31 85.21
Tools
Get Citation
Copy Citation Text
Xingtao Ming, Dehong Yang. Building Extraction from Remote Sensing Image Based on Multi-Module[J]. Laser & Optoelectronics Progress, 2024, 61(4): 0428004
Paper Information
Category: Remote Sensing and Sensors
Received: Apr. 23, 2023
Accepted: Jun. 1, 2023
Published Online: Feb. 26, 2024
The Author Email: Dehong Yang (1486097650@qq.com)
CSTR:32186.14.LOP231148
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20