Super-Resolution Enhanced Data for EO Applications and Services – Topic A: Sentinel-2 Super-Resolution

The one-year ESA project “Super-Resolution Enhanced Sentinel-2 Data for EO Applications and Services” is part of the “EO for Civil Security Applications” activity line of the “Development and Exploitation Component” of ESA’s FutureEO-1 programme (Segment-1) with the main objective to expand uptake of civilian EO capabilities within the wider law enforcement community and civil security stakeholders in particular using super-resolution of Sentinel-2 imagery. The project primarily objective is to extend the use of Sentinel-2 data for EO Applications and Services through the enhancement of spatial resolution of Sentinel-2 products with Deep Learning techniques. Technical objectives are to identify, develop and validate efficient Deep Learning algorithms for resolution enhancement, to assess the performance of super-resolution and the impacts on the considered existing applications of Earth monitoring from satellite imagery with respect to operational constraints in end users application chains.

The consortium gathers recognised experts from the relevant fields of expertise to address the different issues in a consistent way: 1) Thales Services Numériques (France, prime contractor): knowledge of operational uptake for space programs with its rich experience on satellite imagery and Deep Learning techniques; 2) MEOSS (France): Expertise of EO applications and services, strengthening the exploitation of Sentinel-2 for representative use cases from the SCOT de Gascogne and the PETRs of the Pays d´Armagnac, Portes de Gascogne and Pays d´Auch; 3) CESBIO (France): Expertise of EO image data: S2, Venµs, Pleiades.

For this project, the following use cases have been selected to evaluate the contribution of the super resolution:

• Mapping and monitoring of water surfaces, related to agricultural activities. The challenge here is to be able to have spatially resolved information with an intra-annual recurrence to trace the spatio-temporal changes in water surfaces and ultimately to have data on the volumes of water available.  
• Agriculture: hedges and grass strips characterisation. The improvement in level of detail linked to the Super-Resolution of Sentinel-2 could allow better detection of linear landscape elements such as hedges, ditches, grass strips which act as barriers against the risks of erosion, pollution and leaching.
• Urban areas: semi-urban land cover. The challenge in this case is to be able to replace the Very High-Resolution data by Super-Resolution Sentinel-2 data while keeping the accuracy and geometry for small object detection or even that of long-shaped objects such as roads, alignments of roads and trees.

In practice, all of these potential use cases will be implemented in pilot territories in France South-Western providing various range of landscapes and natural features to be fully representative at European level.

The added value of the Sentinel-2 Super-Resolution will be assessed both from a qualitative and operational points of views. The methodology and technical approach used for this project is as follows:

• Reference Data for Training. Two types of sensor perfectly suited to this study will be used: Venµs and Pleiades.
• Deep Learning Networks. State-of-the-art deep learning solutions dedicated to Super Resolution will be tested and deployed with the collaboration of image processing experts.
• Test Cases. Each network at each resolution will be assessed in regards to the project and use cases objectives.