Annonce en lien avec l’Action/le Réseau : aucun
Laboratoire/Entreprise : ISTerre, Université Grenoble-Alpes, France
Durée : 2 years
Contact : sophie.giffard@univ-grenoble-alpes.fr
Date limite de publication : 2020-04-01
Contexte :
Image correlation is a powerful tool for retrieving ground displacements in the near-field of earthquakes. Despite a trend towards higher resolution satellite datasets, medium resolution satellite data represents the cheapest and faster solution for retrieving earthquake displacement maps.
Sujet :
n this project, we aim to explore machine learning approaches (employing deep convolution neural networks) to build a new optical correlator, which is both more robust to noise, less biased by small correlation windows, and more precise when interpolating into the sub-pixel domain. We will test this approach on a unique satellite image dataset of upaired Low Resolution (Sentinel-2) and High Resolution (Pleiades) satellite images, acquired at the same time. If we can establish that the non-linear mappings between LR and HR images can be generalized to regions beyond our training dataset, our correlation approach will be valuable for image correlation in general. We would test this approach on the Californian region, specifically on the 2019 Ridgecrest earthquake, with possible interactions with local researchers.
Profil du candidat :
The profile is rather Applied Mathematics with a desirable experience in machine learning/AI and/or advanced image processing. An interest in Earth Sciences is a plus.
Formation et compétences requises :
! Important ! If you are interested in a post-doc (or a PhD) on this topic, please contact quickly james.hollingsworth@univ-grenoble-alpes.fr and sophie.giffard@univ-grenoble-alpes.fr . We would then actively support a CNES application (deadline : April, 1st 2020, notification for the post-doc: June).
Adresse d’emploi :
ISTerre, Université Grenoble-Alpes