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 aim of our research is to use deep learning methods to detect landslides in satellite images. The study distinguishes between sudden landslides caused by intense rainfall and slow, destructive, long-term landslides. Several methods have been implemented: radar-to-optical modality translation for cloud-independent detection, task combination to make translation more relevant, multimodal detection with missing modalities, InSAR dataset creation and neural network explicability are some of the topics covered in this thesis.

The topic of the researches target sliding areas detections from remote sensing images by using deep learning approaches. Both sudden and slow moving phenomena are of interest in the thesis requiring the implementation of several deep learning techniques. Among them, the major contributions lie in radar-optical modality translation for weather robust detections, leveraging land-cover classification to increase translation reliability, apply multimodal slide detections algorithms within a missing modality context, create a new InSAR dataset or introduce explainability within the networks.