摘要
Cloud removal is essential for pre-processing remote sensing images (RSIs) affected by cloud contamination. Current methods generally adopt thin/thick cloud classifications, overlooking the diversity and variability of natural clouds. Their generalization is limited in real-world scenarios with complex cloud contamination and ground features. The dependence on long-term spectral-temporal data further constrains their applicability. To tackle these challenges, we propose an innovative cloud diffusion model (CDPCR) for more effective cloud removal in RSIs. It realizes natural cloud contamination diffusion and employs an improved sampling algorithm to capture more high-frequency ground details during the generative process. Through reciprocal transitions and iterative learning, it progressively removes complex cloud interference and achieves realistic sequence-to-sequence reconstructions without cloud masks while mitigating the reliance on clean temporal data. Extensive experiments across different datasets demonstrate its effectiveness in diverse and complex cloudy scenarios, outperforming State-of-the-Art methods in qualitative and quantitative evaluations.
| 源语言 | 英语 |
|---|---|
| 主期刊名 | 2025 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS) |
| 出版商 | IEEE; IEEE Circuits and Systems Society (CAS); IEEE Communications Society (ComSoc); IEEE Computer Society; IEEE Signal Processing Society |
| 页 | 6123-6127 |
| 页数 | 5 |
| ISBN(电子版) | 979-8-3315-0810-4 |
| ISBN(印刷版) | 979-8-3315-0811-1 |
| DOI | |
| 出版状态 | 已出版 - 3 8月 2025 |
丛书
| 姓名 | IEEE International Symposium on Geoscience and Remote Sensing IGARSS |
|---|---|
| ISSN(印刷版) | 2153-6996 |
基金资助
This work was supported in part by the National Key Research and Development Program of China under Grant 2022YFB3903501, in part by the GFJG-2024 Annual MultiSpectrum Laboratory Fund under Grant 202410487202.
关键词
- Cloud contamination
- cloud removal
- diffusion model
- image reconstruction
- improved sampling
收录
- CPCI-S
学术指纹
探究 'CLOUD MEETS DIFFUSION: PROGRESSIVE CLOUD REMOVAL FOR OPTICAL REMOTE SENSING IMAGES' 的科研主题。它们共同构成独一无二的学术指纹。引用此
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver