An Adaptive Thin Cloud Removal Method for Mitigating Bright Surface Interference

A thin cloud thickness map (TCTM) is crucial for thin cloud removal tasks, but accurately estimating the TCTM can be a significant challenge. The TCTM is highly susceptible to buildings, which are unavoidable. However, bright surface extraction in existing methods is susceptible to thin clouds, affecting the estimation results for the TCTM. In this article, we propose an adaptive thin cloud removal method for mitigating bright surface interference. Initially, the TCTM is estimated based on the dark pixel search method. The latent low-rank representation method is then employed to decompose the TCTM, resulting in a low-rank optimized TCTM. We also propose a new short-wave infrared-based building index, which effectively extracts building regions in images that are affected by thin clouds, while remaining largely unaffected by them. Based on this index, an adaptive building suppression weight matrix is constructed to reduce the TCTM for building regions covered by thin clouds. For buildings without cloud cover, the inverse proportional averaging method is used to obtain a fixed weight to suppress the TCTM. Finally, the aerosol thickness in the TCTM is excluded, and the thin cloud reflectivity components for each band are resolved through linear relationships combined with scattering laws, achieving thin cloud removal in single remote sensing images. The experimental results demonstrate that the proposed method can accurately extract the TCTM in various scenarios. The thin cloud removal results exhibit high color and structural consistency with the reference images and achieve optimal spectral fidelity.