Researching | Degradation remote sensing image quality enhancement based on frequency-domain-spatial-domain hybrid attention

Optics and Precision Engineering, Volume. 32, Issue 19, 2971(2024)

Degradation remote sensing image quality enhancement based on frequency-domain-spatial-domain hybrid attention

Hua WEI1... Xiongxin TANG1, Haitao NIE2,*, Jing WANG1, Hanxiang YANG1, Yuanyuan XIA1 and Fanjiang XU1 |Show fewer author(s)

Author Affiliations

¹Laboratory of Science and Technology on Integrated Information System， Institute of Software， Chinese Academy of Sciences， Beijing0090， China

²Changchun Institute of Optics， Fine Mechanics and Physics， Chinese Academy of Sciences， Changchun130033， China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (15)

References (43)

Paper Information

Figures & Tables(15)

Fig. 1. Degradation process of remote sensing image

Download full size

View in Article

Fig. 2. Example of remote sensing degraded images

Download full size

View in Article

Fig. 3. Architecture of LFSNet network

Download full size

View in Article

Fig. 4. Frequency domain feature residual module

Download full size

View in Article

Fig. 5. Structure of mixed attention feature extraction module

Download full size

View in Article

Fig. 6. Five different degradation kernels and corresponding degradation images

Download full size

View in Article

Fig. 7. Comparison of restoration visualization in mountainous areas

Download full size

View in Article

Fig. 8. Comparison of restoration visualization in farmland areas

Download full size

View in Article

Fig. 9. Comparison of restoration visualization in village areas

Download full size

View in Article

Fig. 10. Comparison of restoration visualization in port areas

Download full size

View in Article

Fig. 11. MTF curve comparison

Download full size

View in Article

Table 1. Statistics of simulation dataset
View table
View in Article
Table 1. Statistics of simulation dataset
数据集图片数量
训练集 10 000
验证集 15 00

Table 2. NIQE values of different algorithms on high resolution GaoFen-2 images

View in Article

Table 2. NIQE values of different algorithms on high resolution GaoFen-2 images

方法	区域1：山地	区域2：农田	区域3：村落	区域4：港口
原始图像	11.82	9.37	9.30	13.64
Pan等^［11］	9.70	18.88	19.20	26.01
Hu等^［32］	11.54	13.66	12.05	16.19
Chen等^［14］	9.75	15.12	11.37	13.77
Chang等^［31］	10.56	8.38	9.39	12.07
Cho等^［18］	11.39	9.37	9.30	17.78
本文方法	9.62	8.44	7.62	11.20

Table 3. Average running time， deep learning model parameters， and computational complexity of different algorithms on a single Gaofen-2A PMS image

View in Article

Table 3. Average running time， deep learning model parameters， and computational complexity of different algorithms on a single Gaofen-2A PMS image

方法	参数量（M）	计算量（G）	参考时间
Pan等^［11］	None	None	>48 h
Hu等^［32］	None	None	>24 h
Chen等^［14］	None	None	>24 h
Chang等^［31］	6.91	87.9	142 s
Cho等^［18］	16.11	615.7	55 s
本文方法	0.084	0.68	27 s

Table 4. Ablation experiments between modules of LSF algorithm

View in Article

Table 4. Ablation experiments between modules of LSF algorithm

ARFF模块			DSM模块	频域损失函数	PSNR/dB	SSIM
经典特征残差模块	频域特征残差模块	混合注意模块	DSM模块	频域损失函数	PSNR/dB	SSIM
⎫	⎦	⎦	⎦	⎦	27.23	0.741 1
⎫	⎦	⎫	⎦	⎦	27.52	0.743 4
⎦	⎫	⎦	⎦	⎦	28.19	0.758 9
⎦	⎫	⎦	⎫	⎦	28.23	0.761 3
⎦	⎫	⎫	⎫	⎦	28.41	0.762 4
⎦	⎫	⎫	⎫	⎫	28.52	0.763 1

Tools

Get Citation

Copy Citation Text

Hua WEI, Xiongxin TANG, Haitao NIE, Jing WANG, Hanxiang YANG, Yuanyuan XIA, Fanjiang XU. Degradation remote sensing image quality enhancement based on frequency-domain-spatial-domain hybrid attention[J]. Optics and Precision Engineering, 2024, 32(19): 2971

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: May. 8, 2024

Accepted: --

Published Online: Jan. 9, 2025

The Author Email: NIE Haitao (kelek2@126.com)

DOI:10.37188/OPE.20243219.2971

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology