The industry still has work to do to make data analytics less reliant on human oversight

In the quest to turn imagery into useful data for decision-makers, the geospatial intelligence business has gone a long way. But, according to Scot Currie, who serves as the chief solutions architect at the Earth observation startup BlackSky, there is still a lot of work to be done to make machines capable of extracting insights from raw data.

Currie informed SpaceNews on the margins of 2021 GEOINT Symposium that the challenge today is that machine learning and artificial intelligence (AI) are not yet sophisticated enough to fully automate analytics. Currie, who newly joined BlackSky and previously supervised imagery programmes at National Geospatial-Intelligence Agency, said, “The pacing problem is how soon can we get the machines to perform what people have done over the past 50 years?”

BlackSky, which has offices in Virginia, Herndon, and Seattle, bills itself as a “global surveillance” firm that uses a variety of data sources to spot changes or anomalies. It analyses radar and optical images from satellites, photographs taken by aeroplanes and drones, data from the ground sensors, social media and internet-of-things networks using machine learning algorithms.

The company, which just went public, now runs six satellites in inclined orbits to focus coverage over global hotspots, with a goal of expanding to a 30-satellite constellation in the coming years. The majority of its current business comes from government agencies. Customers now must have trust and belief in the information, which still necessitates a significant amount of human analytic labor, according to Currie.

“Monitoring is not yet totally automated,” he added, adding that the industry is working to enhance the technology. Machines will be the primary users of pixels within the next ten to twenty years, according to Currie, and analytics are going to be less demanding on individuals.

As analysts deal with an avalanche of data, it is going to be a race against the clock. “We’re going to see more imagery,” Currie said. “Analysts are overloaded with visuals, so you need to take a different approach.” Allow the machine to count things, identify if something is wrong, and notify you. Then you can assign verification to a collection asset.” “We’re working to mature the algorithms, to be able to train them to be precise enough with the resolution that we have in our constellations,” he said, referring to BlackSky and other industry players. “Are there going to be any false positives?” Yes. Humans, too, make mistakes.”

Leave a comment

Your email address will not be published. Required fields are marked *