Optics and Precision Engineering, Volume. 29, Issue 9, 2222(2021)
Remote sensing image feature extraction and classification based on contrastive learning method
Figures & Tables(13)
Fig. 5. Example images from the four remote sensing scene classification datasets
Fig. 7. Color histograms for the studied data augmentation operators
Table 1. Comparison of the four different remote sensing scene datasets
View tableView in ArticleTable 1. Comparison of the four different remote sensing scene datasets
数据集 类别数 每类样
本数
样本总数 分辨率/m NWPU-RESISC45 45 700 31 500 0.3~30 EuroSAT 10 2 000~3 000 27 000 10~30 UC Merced 21 100 2 100 0.3 SIRI-WHU 12 200 2 400 2
Table 2. Evaluation of the effectiveness of the studied data augmentation operators
View tableView in ArticleTable 2. Evaluation of the effectiveness of the studied data augmentation operators
数据增强方法 CKA
指数
分类准确率（%） Horizontal Flip 0.935 1 63.45 Color Jitter 0.680 3 73.83 Grayscale 0.933 2 67.97 Horizontal Flip+Color Jitter 0.672 7 77.73 Horizontal Flip+Grayscale 0.929 4 66.02 Color Jitter+Grayscale 0.674 6 76.95 Horizontal Flip+Color Jitter+Grayscale 0.670 1 78.13
Table 3. Evaluation of the effectiveness of different AP structures
View tableView in ArticleTable 3. Evaluation of the effectiveness of different AP structures
AP算子深度值 准确率（%） AP2-2 33.59 AP2-3 72.66 AP2-4 78.52 AP2-5 75.78
Table 4. Classification accuracy of different methods
View tableView in ArticleTable 4. Classification accuracy of different methods
方法 NWPU-RESISC45 EuroSAT UC Merced SIRI-WHU 训练使用样本数 5% 20% 5% 20% 5% 20% 5% 20% ResNet18 from scratch 21.54 23.36 56.67 60.11 31.04 43.33 29.55 46.31 ImageNet pre-trained ResNet18 52.31 58.97 82.85 85.45 56.81 77.52 67.09 78.91 SimSiam 73.14 77.08 85.67 88.15 42.59 60.51 50.5 63.17 APCL 73.37 77.57 85.01 87.70 46.98 60.52 56.43 65.83
Table 5. Classification time of different methods
View tableView in ArticleTable 5. Classification time of different methods
方法 NWPU-RESISC45 EuroSAT UC Merced SIRI-WHU 训练使用样本数 5% 20% 5% 20% 5% 20% 5% 20% ResNet18 from scratch 0.38 0.37 0.11 0.11 0.06 0.05 0.06 0.06 ImageNet pre-trained ResNet18 0.38 0.37 0.11 0.11 0.05 0.05 0.04 0.04 SimSiam 4.39 4.39 3.57 3.56 0.53 0.53 0.54 0.54 APCL 4.51 4.50 3.58 3.58 0.54 0.54 0.53 0.53
Tools
Get Citation
Copy Citation Text
Xiao-dong MU, Kun BAI, Xuan-ang YOU, Yong-qing ZHU, Xue-bing CHEN. Remote sensing image feature extraction and classification based on contrastive learning method[J]. Optics and Precision Engineering, 2021, 29(9): 2222
Paper Information
Category: Information Sciences
Received: Jan. 29, 2021
Accepted: --
Published Online: Nov. 22, 2021
The Author Email: BAI Kun (nudt@foxmail. com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20