Deep Learning Based Thin Cloud Removal Fusing Vegetation Red Edge and Short Wave Infrared Spectral Information for Sentinel-2A Imagery

Thin clouds seriously affect the availability of optical remote sensing images, especially in visible bands. Short-wave infrared (SWIR) bands are less influenced by thin clouds, but usually have lower spatial resolution than (Vis) high images (e.g., Sentinel-2A/B, CBERS04, ZY-1 02D and HJ-1B satellites). Most cloud removal methods do not take advantage spectral information available SWIR bands, which affected to restore background tainted Vis In this paper, we propose CR-MSS, a novel deep learning-based method that takes vegetation red edge (VRE) as inputs addition visible/near (Vis/NIR) order improve Sentinel-2 Contrary some traditional methods, use manually designed rescaling algorithm handle at different resolutions, CR-MSS uses convolutional layers automatically process resolution. has two input/output branches Vis/NIR VRE/SWIR, respectively. Firstly, cloudy down-sampled layer low features, then concatenated with corresponding features extracted from VRE/SWIR Secondly, put into fusion tunnel down-sample fuse Third, decomposition is up-sample decompose fused features. Finally, transpose used feature maps input was trained on 28 real Sentinel-2A image pairs over globe, tested separately eight simulated pairs. The average SSIM values (Structural Similarity Index Measurement) for results all testing were 0.69, 0.71, 0.77, 0.81, respectively, 1.74% higher best baseline method. visual demonstrate can produce more realistic shadow methods.