Antoine BRALET

Topic: Multimodal Deep Learning for the analysis of spatio-temporal dynamics from remote sensing images / Apprentissage multimodal profond et analyse des dynamiques spatio-temporelles par imagerie de télédétection

Abstract: The main objective of the thesis is the recovery of landslide susceptibility / hazard maps. In order to do so, we aim at the implementation of a neural network able to deduce such a map from time series of satellite images of different modalities: Optical (RGB) and SAR. The objective also raises other issues such as change detection or modality translation for enhancement purposes.

The major issue of the PhD consists in the creation of susceptibility / dangerosity maps of landslides. We aim at implementing a neural network able to deduce this kind of map from time series of remote sensing images from different modalities - namely Optical (RGB) and SAR. This objective bring aside issues as change detection and modality translation for improvement purposes.