Answers are grounded in this paper and cite their sources.
Existing cloud detection methods struggle to simultaneously capture long-range dependencies and preserve fine-grained local details, limiting accuracy in complex scenes characterized by intricate terrain and diverse cloud formations. To overcome these challenges, we propose V-MTNet, a novel Visual Mamba-Transformer hybrid network for cloud detection in remote sensing images. V-MTNet employs a hierarchical encoder-decoder structure. The encoder integrates a Visual Mamba-Transformer mixer (V-MT), designed to leverage the efficient long-range modeling of Mamba and the superior local detail extraction of Transformer. The decoder integrates the features from the V-MT mixer and the multiscale fusion skip connections. These skip connections are enhanced by spatial-channel dual attention to bridge semantic gaps between the feature maps of the encoder and decoder and compensate for the loss of spatial detail in deep feature maps caused by downsampling. Comprehensive experiments conducted on multispectral remote sensing image datasets from Landsat-8, Sentinel-2, and GaoFen-1 satellites demonstrate that V-MTNet outperforms the state-of-the-art methods and achieves the best mIoU on three datasets: 0.9170, 0.9013, and 0.8910, respectively.
No takes yet. Share an insight, caveat, or question.
Cang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: