AI Finds Nearly 30,000 Unknown Asteroids in Existing Telescope Photos

A graphical depiction of space showing numerous green dots representing space debris orbiting around earth, with earth at the center enclosed by concentric circular orbits.
Each green dot in this image is one of the asteroids that data scientists located with their new AI algorithm. The asteroids were found in existing telescope images from NOIRLab.

Artificial intelligence is used for a lot more generating fake images, scientists also use it to locate patterns in data that would otherwise have been impossible to see. Astronomers also use AI to improve image quality and analyze space photos. Some AI may even help save Earth from a catastrophic asteroid.

Although scientists have spent considerable time searching for asteroids in space, including in the solar system, some have proven very challenging to find. A new AI algorithm has located about 27,500 asteroids hiding in plain sight in existing telescope photos.

As Space reports, the AI algorithm helps scientists track asteroids, some of which could someday collide with Earth.

Many of the newly located asteroids are in the asteroid belt between Mars and Jupiter. Scientists have already found more than 1.3 million asteroids in this part of the solar system, but somehow, nearly 30,000 of them had yet to be found despite being in old telephoto photos.

Of these asteroids, found over about five weeks, 150 have paths that could lead them to Earth’s orbit, although none seem destined to hit the planet. Nonetheless, given that an asteroid can cause significant damage to Earth — sorry, dinosaurs — advance notice is essential. It is a big part of ongoing space science, including part of the mission statement for the 3,200-megapixel LSST PetaPixel recently saw.

The new asteroid-tracking AI algorithm, called Tracklet-less Heliocentric Orbit Recovery, or THOR, looked at more than 400,000 archival images of the night sky maintained by the National Optical-Infrared Astronomy Research Laboratory (NOIRLab). The algorithm needs five observations over 30 days of a specific part of the sky to do its thing, and it has been trained to analyze as many as 1.7 billion light dots in just one image per Space.

The scientists scaled THOR using Google Cloud, which is why the team was featured in the Google video above. The approach is also transferrable to other datasets.

“Not only can we find asteroids in datasets that were never meant for it, but we can make every other telescope in the world better at finding asteroids,” says Ed Lu, executive director of the Asteroid Institute. “It’s a change in how astronomy is done.”

It’s a change that may one day protect Earth from catastrophe.


Image credits: B612 Asteroid Institute / University of Washington DiRAC Institute / OpenSpace Project

Discussion