A full-resolution training framework for Sentinel-2 image fusion

TL;DR

This paper introduces an unsupervised deep learning framework for Sentinel-2 image super-resolution that fuses 10-m and 20-m bands without needing resolution downgrade, showing promising results compared to supervised methods.

Contribution

It proposes a novel unsupervised training scheme with a cycle-consistency loss for Sentinel-2 image fusion, eliminating the need for resolution downgrade in training data.

Findings

Promising results compared to supervised approaches

Avoids resolution downgrade process in training

Framework ascribes to multi-resolution analysis methods

Abstract

This work presents a new unsupervised framework for training deep learning models for super-resolution of Sentinel-2 images by fusion of its 10-m and 20-m bands. The proposed scheme avoids the resolution downgrade process needed to generate training data in the supervised case. On the other hand, a proper loss that accounts for cycle-consistency between the network prediction and the input components to be fused is proposed. Despite its unsupervised nature, in our preliminary experiments the proposed scheme has shown promising results in comparison to the supervised approach. Besides, by construction of the proposed loss, the resulting trained network can be ascribed to the class of multi-resolution analysis methods.