Matching radar and optical satellite images has never been easier
The neural network predicts areas of possible match across data modalities © Hughes et al., J. ISPRS 2020
First publication of the ECEO lab, in the Journal of the ISPRS!
In this study, we developed a methodology to find corresponding points in both very high resolution optical images and synthetic aperture radar ones of the same region. Each sensor has its strenghts and flaws, and being able to use them concurrently is a great asset for environmental monitoring. But finding matching points is very difficult, since the images correspond to completely different signals issued from different wavelenghts of light and acquired with technologies based on different philosophies.
Using AI and deep learning, we train a machine learning model able to find sparse matching points across images and provide a new look to the problem of multi-sensor image matching.
A collaboration between EPFL, WUR, TUMunich and Université de Paris.
The paper is accessible open access at the Jounral of the ISPRS.
Hughes, Marcos, Lobry, Tuia, Schmitt, A deep learning framework for matching of SAR and optical imagery, J. ISPRS, 2020.