Authors:
Théo Archambault
1
;
2
;
Arthur Filoche
1
;
3
;
Anastase Charantonis
4
;
5
and
Dominique Béréziat
1
Affiliations:
1
Sorbonne Université, CNRS, LIP6, Paris, France
;
2
Sorbonne Université, LOCEAN, Paris, France
;
3
UWA, Perth, Australia
;
4
ENSIIE, LaMME, Evry, France
;
5
Inria, Paris, France
Keyword(s):
Image Inverse Problems, Deep Neural Network, Spatiotemporal Inpainting, Multi-Variate Observations, Transfer Learning, Satellite Remote Sensing.
Abstract:
The ocean is observed through satellites measuring physical data of various natures. Among them, Sea Surface Height (SSH) and Sea Surface Temperature (SST) are physically linked data involving different remote sensing technologies and therefore different image inverse problems. In this work, we propose to use an Attention-based Encoder-Decoder to perform the inpainting of the SSH, using the SST as contextual information. We propose to pre-train this neural network on a realistic twin experiment of the observing system and to fine-tune it in an unsupervised manner on real-world observations. We show the interest of this strategy by comparing it to existing methods. Our training methodology achieves state-of-the-art performances, and we report a decrease of 25% in error compared to the most widely used interpolations product.