Thin cloud detection method in thin cloud scenarios based on class center residual attention

作者：Mingjie Zhang, Jiajie He, Shilin Zhou, Pu Wang · 年份：2026 · DOI：10.1117/12.3114962 · 被引用次数：1 · 研究领域：Atmospheric aerosols and clouds、Remote Sensing in Agriculture、Solar Radiation and Photovoltaics

The semi-transparent nature of thin clouds leads to similar gray levels between cloud layers and underlying surfaces, resulting in intra-class feature dispersion—thin clouds exhibit different characteristics due to differences in spectral reflection properties of underlying surfaces, making thin cloud detection one of the most challenging tasks in remote sensing image cloud processing. To address these issues, this chapter proposes a thin cloud detection framework based on class center residual attention mechanism, breaking through the over-reliance of traditional models on local features. This method constructs invariant feature representations of thin clouds in multi-underlying surface scenarios by stripping off underlying surface interference components through class center residual calculation and focusing on radiation difference-sensitive areas using attention mechanism. Experiments show that this method improves the thin cloud IoU to 85.93% on the Landsat-8 thin cloud dataset, providing a new technical paradigm for refined processing of remote sensing images in cloudy areas.