Depth Separable architecture for Sentinel-5P Super-Resolution

TL;DR

This paper introduces a novel depth separable convolution-based super-resolution model tailored for Sentinel-5P satellite data, significantly enhancing spatial resolution and detail capture in hyperspectral atmospheric measurements.

Contribution

The work presents a new super-resolution model using Depth Separable Convolution architecture specifically designed for Sentinel-5P hyperspectral data, outperforming existing methods.

Findings

Model outperforms existing super-resolution techniques on most spectral bands.

Effective exploitation of cross-channel correlations improves spatial detail.

Enhances the potential for precise atmospheric analysis and remote sensing applications.

Abstract

Sentinel-5P (S5P) satellite provides atmospheric measurements for air quality and climate monitoring. While the S5P satellite offers rich spectral resolution, it inherits physical limitations that restricts its spatial resolution. Super-resolution (SR) techniques can overcome these limitations and enhance the spatial resolution of S5P data. In this work, we introduce a novel SR model specifically designed for S5P data that have eight spectral bands with around 500 channels for each band. Our proposed S5-DSCR model relies on Depth Separable Convolution (DSC) architecture to effectively perform spatial SR by exploiting cross-channel correlations. Quantitative evaluation demonstrates that our model outperforms existing methods for the majority of the spectral bands. This work highlights the potential of leveraging DSC architecture to address the challenges of hyperspectral SR. Our model allows for capturing fine details necessary for precise analysis and paves the way for advancements in air quality monitoring as well as remote sensing applications.