WASHINGTON — Days after a Chinese spy balloon was shot down by the U.S. military off the coast of South Carolina, AI startup Synthetaic backtracked the balloon’s trajectory across the continental United States using Earth imaging data from Planet Labs satellites.
Synthetaic founder Corey Jaskolski said this is a novel way to use commercial satellite imagery to track aerial vehicles that might not be seen by radar or by ground observers. It’s also another way to exploit open-source data to independently confirm where the Chinese balloon or other objects came from, he said. “The map we put together overlays our sightings and all the social media reports of where the balloon was.”
With the aid of an AI image recognition tool, Jaskolski on Feb. 11 found the balloon in Planet’s satellite data relatively quickly, he told SpaceNews Feb. 17.
He first saw the balloon in the satellite data over the Atlantic about two hours before an Air Force fighter jet took it down Feb. 4. “We then started tracing backwards across the United States and got a total of four hits: two in South Carolina, one in Missouri, and one in Canada.”
“The hit that we got in Canada was over snow and ice, and we were still able to pick up the balloon,” he said. “We were able to take measurements from the pixels and validate it was about 148 feet in diameter and flying at 58,000 feet. Once we had those calculations, we knew for sure it was the balloon. As soon as we found it, we had a signature that we could use in RAIC and we could start working our way backwards.”
RAIC is the company’s AI tool — short for Rapid Automatic Image Categorization. It’s different from most AI systems, he said, because it doesn’t require pre-trained models or extensive image labeling, and quickly analyzes unsupervised data.
Faster way to analyze
“All of the AI being developed today is data hungry,” said Jaskolski, an MIT engineer and explorer who founded Synthetaic in 2019.
“Feeding AI with high-quality labeled data is expensive and time-consuming, and the number one thing keeping us from applying AI more widely and more efficiently,” he added. “In a national security emergency, spending months labeling data is not really a luxury we have.”
The company has worked with the nonprofit Climate TRACE to identify the location of concentrated animal feeding operations, and with National Geographic to help track rare species endangered by poaching.
Jaskolski said this AI system can be set up for daily monitoring to track any aerial objects, and he’s not aware of any government programs that have been doing this from space.
President Biden on Feb. 16 said he ordered the U.S. military to shoot down three aerial objects last weekend “out of an abundance of caution” and the objects turned out to be harmless.
The ability to trace these objects to their place of origin would help answer a lot of questions, Jaskolski said. “If it was launched from the corporate headquarters of a weather company, that would suggest it was a weather balloon. If it was launched from a military base, it would likely be something else.”
A tool like RAIC runs on geospatial data, full motion video and imagery, he added, “and it allows people to answer questions like this really, really quickly.”
The balloon incident is interesting and the problem du jour, Jaskolski said, but other situations may come up “where we will not have a pre-trained AI model for” and there will not be enough time to train analysts.
Synthetaic has not yet completely traced the spy balloon’s trajectory to the point of origin. “We are currently running RAIC across more satellite image data over Alaska and Asia, but we have not yet gone beyond tracing it back to the northwestern-most point on the map in Canada, just north of Spokane, Washington,” he said.
According to U.S. intelligence officials, the balloon’s home base was near China’s south coast. Others said it originated in Central China. And China admitted it was theirs, Jaskolski said, “but with RAIC we hope to verify its exact origin soon.”