Research on Thin Cloud Removal Based on Generative Adversarial Network with CBAM and Multi-Scale Attention

Yang Wang; Guokun Chen; Xingwu Duan; Qingke Wen; Jiatian Li; Zhen Zhang

doi:10.3788/AOS250568

Acta Optica Sinica, Volume. 45, Issue 12, 1210001(2025)

Research on Thin Cloud Removal Based on Generative Adversarial Network with CBAM and Multi-Scale Attention

Yang Wang¹, Guokun Chen^1,2,3、*, Xingwu Duan^4,5, Qingke Wen^6,7, Jiatian Li¹, and Zhen Zhang^1,2,3

Author Affiliations

¹Faculty of Land Resources Engineering, Kunming University of Science and Technology, Kunming 650093, Yunnan , China

²Yunnan Key Laboratory of Quantitative Remote Sensing, Kunming 650093, Yunnan , China

³Yunnan International Joint Laboratory for Integrated Sky-Ground Intelligent Monitoring of Mountain Hazards, Kunming 650093, Yunnan , China

⁴Institute of International Rivers and Eco-Security, Yunnan University, Kunming 650500, Yunnan , China

⁵Yunnan Key Laboratory of Soil Erosion Prevention and Green Development, Yunnan University, Kunming 650500, Yunnan , China

⁶Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

⁷National Engineering Research Center for Geomatics, Beijing 100101, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(18)

Fig. 1. Network structure of the thin cloud removal method based on CBAM and multi-scale attention generative adversarial network

Download full size

Fig. 2. Structure diagram of the generator

Download full size

Fig. 3. Structure diagram of the discriminator

Download full size

Fig. 4. Structure diagram of the CBAM. (a) Channel attention module; (b) spatial attention module

Download full size

Fig. 5. Multi-scale attention module

Download full size

Fig. 6. Examples of the RICE remote sensing cloud removal dataset

Download full size

Fig. 7. Examples of a remote sensing cloud removal dataset based on Sentinel-2

Download full size

Fig. 8. Generated results based on the RICE1 dataset

Download full size

Fig. 9. Generated results based on the Sentinel-2 dataset

Download full size

Fig. 10. Qualitative analysis of the pixel space of the RICE1 dataset

Download full size

Fig. 11. Qualitative analysis of pixel space based on the Sentinel-2 dataset

Download full size

Fig. 12. Average brightness of the RGB channels of the images generated by each model based on the RICE1 dataset

Download full size

Fig. 13. Average brightness of the RGB channels of the images generated by each model based on the Sentinel-2 dataset

Download full size

Fig. 14. Processing results of the ablation experiment on the RICE1 dataset. (a) Original cloud image; (b)(c) images generated by combining CBAM and multi-scale attention and the corresponding attention heat map; (d) (e) images generated by adding multi-scale attention and the corresponding attention heat map; (f)(g) images generated by adding CBAM and the corresponding attention heat map

Download full size

Table 1. Comparison of quantitative results of different models on the RICE1 dataset
View table
Table 1. Comparison of quantitative results of different models on the RICE1 dataset
Model PSNR /dB SSIM
CGAN 28.218 0.872
SpA-GAN 30.048 0.889
C²PNet 20.202 0.869
Haze Removal 23.678 0.827
FFA-Net 16.382 0.772
Ours 31.321 0.894

Table 2. Comparison of quantitative results of different models on the Sentinel-2 dataset
View table
Table 2. Comparison of quantitative results of different models on the Sentinel-2 dataset
Model PSNR /dB SSIM
CGAN 26.806 0.822
SpA-GAN 28.905 0.843
C²PNet 23.488 0.816
Haze Removal 20.583 0.797
FFA-Net 12.951 0.569
Ours 29.668 0.869

Table 3. Comparison of model parameters
View table
Table 3. Comparison of model parameters
Model Model parameter quantity
FFA-Net 4455913
C²PNet 7169327
CGAN 7065028
SpA-GAN 2982245
Ours 2956387

Table 4. Comparison of quantitative results of ablation experiments on the RICE1 dataset
View table
Table 4. Comparison of quantitative results of ablation experiments on the RICE1 dataset
Condition PSNR /dB SSIM
Multi-scale attention 30.245 0.843
CBAM 28.749 0.803
Multi-scale+CBAM 31.321 0.894

Tools

Get Citation

Copy Citation Text

Yang Wang, Guokun Chen, Xingwu Duan, Qingke Wen, Jiatian Li, Zhen Zhang. Research on Thin Cloud Removal Based on Generative Adversarial Network with CBAM and Multi-Scale Attention[J]. Acta Optica Sinica, 2025, 45(12): 1210001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Feb. 5, 2025

Accepted: Apr. 24, 2025

Published Online: Jun. 23, 2025

The Author Email: Guokun Chen (chengk@radi.ac.cn)

DOI:10.3788/AOS250568

CSTR:32393.14.AOS250568

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of quantitative results of different models on the RICE1 dataset

Table 1. Comparison of quantitative results of different models on the RICE1 dataset

Table 2. Comparison of quantitative results of different models on the Sentinel-2 dataset

Table 2. Comparison of quantitative results of different models on the Sentinel-2 dataset

Table 3. Comparison of model parameters

Table 3. Comparison of model parameters

Table 4. Comparison of quantitative results of ablation experiments on the RICE1 dataset

Table 4. Comparison of quantitative results of ablation experiments on the RICE1 dataset