RepreSent: Non-supervised Representation Learning for Sentinels
Supervised Deep Learning techniques in EO often depend on labelled data. It would be expensive and time-consuming to obtain such labels for the massive amount of EO data that the Sentinel satellites of the Copernicus program have been gathering. In order to effectively exploit this abundant pool of data, the project “RepreSent” will therefore harness the power of non-supervised learning. Together with the subcontractors EPFL, VTT, and e-GEOS, we will design, implement and validate appropriate AI algorithms using five use cases:
(1) Forest disturbance monitoring, (2) automated land cover mapping, (3) anomaly detection in long time series data, (4) cloud detection and removal, and (5) forest biomass estimation.
The resulting datasets will be made freely available to the community. We will further make sure to reach out to relevant networks and startups to raise awareness and promote our results.
We will also engage in Public Outreach, for example via ESA’s web stories and Social Media channels.
Links
