Thin Cloud Removal Based on the Sorted Slow Feature Analysis for Multispectral Remotely Sensed Imagery

This study proposes a novel thin cloud removal algorithm based on sorted slow feature analysis (s-SFA), exploiting the spatial slow variation feature of thin clouds. The algorithm is applied to Landsat-8 Operational Land Imager (OLI) data. The effectiveness of the proposed algorithm is assessed using both simulated and real cloud-affected data. Results demonstrate significant improvements in R
²
and peak signal-to-noise ratio (PSNR) values for the simulated cloud-affected data, indicating efficient thin cloud removal. Further validation using real Landsat-8 data confirms the algorithm's effectiveness over various land use and land cover types while preserving satisfactory ground features in cloud-free regions.