12 mars 2021
Offre de thèse en intelligence artificielle et télédétection
Catégorie : Doctorant
Phd Position: Self-supervised representation learning from multi-domain data for land cover change detection
This 3-year PhD program will be funded by the ANR- DeepChange projet. DeepChange is an encouraging interdisciplinary project focused on investigating Deep Generative models to detect land cover changes by using satellite images. The project aims at detecting land cover conversion changes by comparing multivariate satellite image sequences acquired on different time periods. The exploitation of satellite temporal sequences is not straighforward since they are non-stationary,not dense, multi-variate and have temporal gaps. The main novelty of the project is the develop-
ment of deep learning methodologies to build new spatio-temporal representations uncovering the underlying data distributions of complex time series exhibiting spatial dependencies. Mining valuable knowledge from spatio-temporal data is critically important to many real world applications and DeepChange project proposes cutting-edge research topics as self-supervised, cross-domain and task-guided representation learning.
This PhD aims to investigate new change detection methodologies to map abrupt land-cover conversion changes by exploiting new Satellite Images Times Series (SITS). Land cover conversion changes are defined as landscape transitions between two time periods. To detect and characterize land-cover changes, this PhD aims to discover useful semantic low-dimensional representations from SITS. The PhD work is dedicated to develop new methodologies from a domain adaptation perspective. Cross-domain self-supervised representation learning strategies relying on Deep Generative models will be investigated. First, the multi-view scene analysis will be performed by studying multi-domain image sequences. The term multi-domain refers to a pair of SITS acquired over the same scene but on different periods. In a second stage, the PhD will explore the multi-modal challenge that involves SITS acquired by different sensors.
Requirements: The candidate must have a background in statistical signal and image processing, machine learning and scientific programming (Python, C/C++). A good knowledge of English is required.
Contact: Candidates should send an e-mail to email@example.com and firstname.lastname@example.org
• Full CV,
• Motivation letter,
• Contact information for 2 references, and/or recommendation letter.
Application is open until the position is fulfilled.
(c) GdR 720 ISIS - CNRS - 2011-2020.