Synthetic data helps train algorithms to spot rare objects

DENVER — Despite ever-expanding geospatial databases, some objects like specific submarines are rarely spotted in satellite images. In these cases, computer-generated imagery can help.

“Subject matter experts to identify objects in scenarios are always very important,” said Mark Munsell, deputy director of digital data and information for the National Geospatial-Intelligence Agency. SpaceNews. “We anticipate that they will be supplemented with synthetic in the future.”

Computer vision models only work after ingesting a lot of data about various objects, which is not always available.

“Synthetic data is, I believe, the future,” Kevin O’Brien, CEO of Orbital Insight. “Being able to look at what we call rare items, specific types of missile launchers or other things that can be used for military purposes, is going to benefit our end customers like NGA, but I think it’s also going to benefit industry in general.”

Orbital Insight is working under an NGA Phase 2 Small Business Innovation Research contract to explore the use of synthetically generated data to train computer vision models to detect new objects in imagery electro-optical satellite.

At the GEOINT 2022 Symposium, L3Harris Technologies showcased its ability to create synthetic formation data for both electro-optical and synthetic aperture radar data.

“We have a proven synthetic training data capability that provides an alternative solution for those times when you don’t have enough live images,” said William Rorrer, director of business development and product manager at L3Harris.

For example, L3Harris shows how the company creates synthetic images of fighter jets in different contexts, in various weather conditions and at different angles of view and sunlight.

“We focused on synthetic data accuracy,” said Stacey Casella, L3Harris senior manager for geospatial processing and analysis. “As long as you have that precision, training AI algorithms can be robust.”

L3Harris is among companies that have won base order agreements, similar to indefinite-delivery, indefinite-quantity awards, from the Department of Defense’s Joint Artificial Intelligence Center Data Readiness for Artificial Intelligence Development, a five-year program with a cap of $242 million aimed at leveraging commercial capabilities to address technical challenges.

L3Harris’ synthetic data work builds on the company’s experience as a sensor manufacturer.

“We have a history of creating fake data for things like commercial satellite sensor studies,” Rorrer said. “We want to understand what that image will look like and have a very precise understanding of the scans before putting a payload into orbit.”

Synthetic formation data was a popular topic at the GEOINT Symposium. Conferences on the subject have been presented by CACI International, Rendered.ai, Riverside Research, Scale AI and Visimo.

Sharon D. Cole