On The Role of Alias and Band-Shift for Sentinel-2 Super-Resolution

TL;DR

This paper investigates how Sentinel-2's unique sensor characteristics, like inter-band shift and alias, can be exploited by deep learning for effective single-image super-resolution without hallucinating details.

Contribution

It introduces a super-resolution method leveraging Sentinel-2's sensor properties and demonstrates its effectiveness with a new dataset and simple training approach.

Findings

Deep-learning models can recover fine details thanks to Sentinel-2's sensor features.

Training with a simple L1 loss yields realistic super-resolution results.

A new Sentinel-2/PlanetScope dataset supports evaluation of SR methods.

Abstract

In this work, we study the problem of single-image super-resolution (SISR) of Sentinel-2 imagery. We show that thanks to its unique sensor specification, namely the inter-band shift and alias, that deep-learning methods are able to recover fine details. By training a model using a simple $L_1$ loss, results are free of hallucinated details. For this study, we build a dataset of pairs of images Sentinel-2/PlanetScope to train and evaluate our super-resolution (SR) model.