Remote Sensing Scene Classification Based on Local Selection Vision Transformer

Kai Yang; Xiaoqiang Lu

doi:10.3788/LOP230539

Laser & Optoelectronics Progress, Volume. 60, Issue 22, 2228005(2023)

Remote Sensing Scene Classification Based on Local Selection Vision Transformer

Kai Yang^1,2 and Xiaoqiang Lu^1、*

¹Key Laboratory of Spectral Imaging Technology, Xi'an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences, Xi'an 710119, Shaanxi , China

²University of Chinese Academy of Sciences, Beijing 100049, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(8)

Fig. 1. Remote-sensing image examples

Download full size

Fig. 2. Network structure of LSViT

Download full size

Table 1. Comparison results of different Patch divisions
View table
Table 1. Comparison results of different Patch divisions
Patch division OA /% Training time /min
Non-overlapping 95.21 59.82
Overlapping 95.33 62.91

Table 2. Comparison results of different step sizes
View table
Table 2. Comparison results of different step sizes
Step size 8 10 12 14 16
OA /% 94.73 94.90 95.12 95.33 95.21

Table 3. Impact result of the local selection module
View table
Table 3. Impact result of the local selection module
Method OA /%
ViT+overlapping 95.01
LSViT 95.33

Table 4. Comparison results with the CNN model
View table
Table 4. Comparison results with the CNN model
Method OA /%
VGG16 88.18
GoogleNet 92.88
ResNet-18 92.96
ResNet-50 93.24
ViT 94.90
LSViT 95.33

Table 5. Comparison with state-of-the-art methods on AID dataset

View table

Table 5. Comparison with state-of-the-art methods on AID dataset

Method	Backbone	OA /%
Method	Backbone	20% training data	50% training data
GBNet^［19］	VGG16	92.20±0.23	95.48±0.12
IBCNN^［20］	MobileNet-V2	94.23±0.16	96.57±0.28
ACR-MLFF^［6］	ResNet-50	92.73±0.12	95.06±0.33
SAFF^［21］	VGG-VD16	90.25±0.29	93.83±0.28
Ji et al.+VGG-VD16^［7］	VGG-VD16	94.75±0.23	96.93±0.16
ResNet50+EAM^［22］	ResNet50	93.64±0.25	96.62±0.13
CNN-GCN^［23］	VGG16	94.93±0.31	96.89±0.10
SKAL^［24］	ResNet18	94.38±0.10	96.76±0.20
LSViT	ViT	95.33±0.05	97.55±0.12

Table 6. Comparison with state-of-the-art methods on NWPU-RESISC45 dataset

View table

Table 6. Comparison with state-of-the-art methods on NWPU-RESISC45 dataset

Method	Backbone	OA /%
Method	Backbone	10% training data	20% training data
IBCNN^［20］	MobileNet-V2	90.49±0.17	93.33±0.21
ACR-MLFF^［6］	ResNet-50	90.01±0.33	92.45±0.20
SAFF^［21］	VGG-VD16	84.38±0.19	87.86±0.14
Ji et al.+VGG-VD16^［7］	VGG-VD16	91.08±0.24	93.49±0.17
ResNet50+EAM^［22］	ResNet50	90.89±0.15	93.51±0.12
CNN-GCN^［23］	VGG16	90.75±0.21	90.92±0.24
SKAL^［24］	ResNet18	90.04±0.15	92.79±0.11
LSViT	ViT	92.41±0.15	94.25±0.25

Tools

Get Citation

Copy Citation Text

Kai Yang, Xiaoqiang Lu. Remote Sensing Scene Classification Based on Local Selection Vision Transformer[J]. Laser & Optoelectronics Progress, 2023, 60(22): 2228005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites