Semantic segmentation of clouds and cloud shadows using state space models

Zhang, Z., Hu, Z. ORCID: https://orcid.org/0000-0001-9994-771X, Xia, M. ORCID: https://orcid.org/0000-0003-4681-9129, Yan, Y. ORCID: https://orcid.org/0000-0002-3609-0496, Zhang, R., Liu, S. and Li, T. ORCID: https://orcid.org/0000-0002-3418-275X (2025) Semantic segmentation of clouds and cloud shadows using state space models. Remote Sensing, 17 (17). 3120. ISSN 2072-4292 doi: 10.3390/rs17173120

Abstract/Summary

In remote sensing image processing, cloud and cloud shadow detection is of great significance, which can solve the problems of cloud occlusion and image distortion, and provide support for multiple fields. However, the traditional convolutional or Transformer models and the existing studies combining the two have some shortcomings, such as insufficient feature fusion, high computational complexity, and difficulty in taking into account local and long-range dependent information extraction. In order to solve these problems, this paper proposes the MCloud model based on Mamba architecture is proposed, which takes advantage of its linear computational complexity to effectively model long-range dependencies and local features through the coordinated work of state space and convolutional support and the Mamba-convolutional fusion module. Experiments show that MCloud have the leading segmentation performance and generalization ability on multiple datasets, and provides more accurate and efficient solutions for cloud and cloud shadow detection.

Item Type	Article
URI	https://centaur.reading.ac.uk/id/eprint/124421
Identification Number/DOI	10.3390/rs17173120
Refereed	Yes
Divisions	Science > School of Mathematical, Physical and Computational Sciences > Department of Computer Science
Publisher	MDPI
Download/View statistics	View download statistics for this item