Optics and Precision Engineering, Volume. 31, Issue 17, 2555(2023)
Partial optimal transport-based domain adaptation for hyperspectral image classification
Figures & Tables(8)
Fig. 1. Framework of proposed domain adaptive model base on partial optimal transport
Fig. 2. Pseudocolor images and corresponding ground truths with color indexes of Pavia dataset
Fig. 3. Pseudocolor images and corresponding ground truths with color indexes of Houston dataset
Fig. 4. Classification maps of PC dataset produced by different methods
Table 1. Land cover classes and numbers of samples in Pavia dataset
View tableView in ArticleTable 1. Land cover classes and numbers of samples in Pavia dataset
类别 样本数量 UP（源域） PC（目标域） 牧场 18 649 3 090 阴影 947 2 863 砖块 3 682 2 685 沥青道路 6 631 9 248 树木 3 064 7 598 沥青 1 330 7 287 裸露土壤 5 029 6 584 总计 39 332 39 355
Table 2. Land cover classes and numbers of samples in Houston dataset
View tableView in ArticleTable 2. Land cover classes and numbers of samples in Houston dataset
类别 样本数量 H13（源域） H18（目标域） 草地 345 1 353 道路 443 6 365 受压草地 365 2 766 水面 285 22 树木 365 2 766 居民建筑 319 5 347 非居民建筑 408 32 459 总计 2 530 53 200
Table 3. Classification performance on OA，AA（%） and Kappa of different target domains
View tableView in ArticleTable 3. Classification performance on OA，AA（%） and Kappa of different target domains
Method UP→PC H13→H18 OA（%） AA（%） K（×100） OA（%） AA（%） K（×100） TSVM 76.03±0.02 72.41±2.22 70.70±3.05 52.03±0.32 62.40±0.40 34.29±0.80 DAN 80.68±0.02 76.18±1.62 76.47±2.34 61.45±2.18 73.47±1.74 47.08±2.36 DANN 83.90±0.05 78.40±0.47 80.41±0.50 62.61±2.10 65.09±3.54 49.33±1.98 ED-DMM-UDA 82.50±0.50 77.80±0.56 78.73±0.54 66.09±1.83 74.48±1.66 53.10±2.18 CDA 89.78±3.03 88.80±5.05 87.71±3.66 68.44±0.77 75.31±1.58 55.65±0.77 CLDA 92.80±0.67 91.83±1.09 91.32±0.81 70.12±2.43 79.07±0.38 57.75±2.49 Ours 93.67 ±0.50 94.17 ±1.06 93.76 ±0.71 70.25 ±2.20 79.89 ±1.06 58.32 ±1.22
Table 4. Ablation study classification results of OA with different transfer tasks
View tableView in ArticleTable 4. Ablation study classification results of OA with different transfer tasks
Method UP→PC H13→H18 baseline 91.46 67.50 +OT 91.96 68.24 +OT+CS 92.85 69.27 +POT（s=0.5）+CS 93.05 69.43 +POT（s-adapt）+CS 93.67 70.25
Tools
Get Citation
Copy Citation Text
Bilin WANG, Shengsheng WANG, Zhe ZHANG. Partial optimal transport-based domain adaptation for hyperspectral image classification[J]. Optics and Precision Engineering, 2023, 31(17): 2555
Paper Information
Category: Information Sciences
Received: Feb. 15, 2023
Accepted: --
Published Online: Oct. 9, 2023
The Author Email: WANG Shengsheng (wss@jlu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20