
Enhancing maritime domain awareness using AI and embeddings
November 10 @ 10:00 am – 11:00 am
CET
The webinar, organised by SatCen, showcases how the use of AI and latent spaces with Earth Observation (EO) data could improve the current practices of Maritime Domain Awareness (MDA).
The study introduces a methodology that, instead of detecting vessels using original Sentinel-1 and -2 imagery, uses highly compressed embeddings derived from TerraMind, a Geospatial Foundational Model (GFM) developed by IBM and ESA. Relying on a dataset of 100,000 balanced embeddings, this work demonstrates that critical MDA tasks can be performed using embeddings that are up to 100x smaller than the original data, with an accuracy aligned to the one obtained with original images. The study will show the application of the methodology on a large scale area, represented by the Mediterranean Sea.
The webinar will collect feedback from the audience, in order to foster discussions on potential applications, align solutions with user needs, and encourage the adoption of the pilot’s outcomes to address real-world scenarios.

