Spatial-Characteristics Driven Transmission Map Estimation for the Haze Removal in Sentinel-2 Spectral Bands Using the S2 China Hazy-MSI Dataset
Balla Pavan Kumar, Aravinth J, Sankaran Rajendran, Arvind Mukundan · IEEE Access · 2025
The Sentinel-2 satellite-based Multi-Spectral Imagery (MSI) is crucial for Earth observations owing to its high spectral and spatial resolution. However, the bands of this satellite MSI may be affected by the haze at various times due to changes in weather conditions. Although there are several existing algorithms for dehazing, their primary focus is on RGB images, which may also result in low brightness, colour saturation, and lower detail outcomes. To address these shortcomings, we propose a novel transmission map that leverages spatial haze characteristics - patch-wise local luminance, patch-wise local intensity max-min difference, and atmospheric light, for improved dehazing performance. Then, a wavelength-dependent transmission adjustment is applied individually to each spectral band of the hazy Sentinel-2 MSI. Subsequently, the dehazed image is evaluated on a per-band basis by leveraging both the estimated atmospheric light and the refined transmission map, utilizing the principles of the atmospheric scattering model. The experimentation is performed on a newly developed dataset called Sentinel-2 China Hazy MSI (S2CH-MSI), which has been developed by simulating various haze patterns for several geographically diverse locations of the haze-prone China region. The results of our proposed work show significant improvements in terms of qualitative and quantitative assessments in comparison to those of the existing haze removal methods. The code for the proposed haze-removal method is available at https://github.com/kumarballapavan/SCTE.