Romain Wenger

Apport des images Sentinel-1&2 et des méthodes d’apprentissage profond pour la cartographie et le suivi des modes d’occupation des sols: La couverture biophysique de la surface terrestre, plus communément appelée occupation des sols, est une composant des Variables Climatiques Essentielles (VCE) depuis plusieurs années. Dans un contexte de changement globale, sa connaissance précise est au cœur de nombreux projets de recherche sur l’environnement. En effet, cette donnée peut être utilisé en entrée de différents modèles, tels que les modèles climatiques et permet d’aider les acteurs locaux des territoires dans leur prise de décision. Les constellations satellitaires actuelles fournissent des images à haute résolution spatiale et temporelle avec de multiples modes d’acquisition (radar, optique etc.). Les nouvelles méthodes d’intelligence artificielle telles que l’apprentissage profond, ont permis d’obtenir des résultats prometteurs et ont accéléré le traitement des données massives. Ainsi, l’objectif de cette thèse est d’évaluer la contribution des capteurs Sentinel-1&2 pour la classification et le suivi de l’occupation et de l’usage des sols et plus particulièrement des tissus urbains.

The size of urban settlements is rapidly increasing worldwide. This sprawl triggers changes in land cover with the consumption of natural areas and affects ecosystems with important ecological, climate, and social transformations. Detecting, mapping, and monitoring the growth and spread of urban areas is therefore important for urban planning, risk analysis, human health, and biodiversity conservation. Satellite images have long been used to map human settlements. The availability of the Sentinel constellation (S2) allows the monitoring of urban sprawl over large areas (e.g., countries) and at high frequency (with possible monthly updates). This massive data stream allows the proposal of new types of urban products at a spatial resolution of 10 meters. In this context, we developed a fully automated and supervised processing chain (URBA-OPT) using open-source libraries and optimized for rapid calculation on high-performance computing clusters. This processing chain has been applied to all of France. The objective is to propose an annual product at 10 m spatial resolution available through the THEIA Data and Services Centre. Our results demonstrate the feasibility and accuracy of producing an annual urban settlement map using this methodology, providing a valuable tool for urban planning and environmental monitoring.

This paper presents MultiSenGE that is a new large scale multimodal and multitemporal benchmark dataset covering one of the biggest administrative region located in the Eastern part of France. MultiSenGE contains 8,157 patches of 256 × 256 pixels for the Sentinel-2 L2A , Sentinel-1 GRD images in VV-VH polarization and a Regional large scale Land Use/Land Cover (LULC) topographic reference database. With MultiSenGE, we contribute to the recents developments towards shared data use and machine learning methods in the field of environmental science. The purpose of this dataset is to propose relevant and easy-access dataset to explore deep learning methods. We use MultiSenGE to evaluate the performance for urban areas using well-known deep learning techniques. These results serve as a baseline for future research on remote sensing applications using the multi-temporal and multimodal aspects of MultiSenGE. With all patches georeferenced at a 10 meters spatial resolution covering the whole Grand-Est Region, MultiSenGE provides an opportunity for environmental benchmark dataset will help to advance data-driven techniques for land use/land cover remote sensing applications.