AI bests humans at mapping the moon
A computer algorithm found about 6,000 lunar craters that hadn't been counted before
BRACE FOR IMPACT Artificial intelligence paints a portrait of the moon that’s even more pockmarked than previously thought.
NASA
Artificial intelligence is helping draw a more detailed map of the moon.
An AI that studied lunar images to learn what craters look like has discovered thousands of new pockmarks on the moon’s surface. This program could also be used to catalog impact scars on other moons or planets, which might improve scientists’ understanding of how various objects roamed our solar system in the past.
The new algorithm is an artificial neural network that attempts to mimic the way the brain processes information. After training on Lunar Reconnaissance Orbiter images covering about one-third of the moon’s surface, the program was shown another third of the lunar landscape. The AI identified 92 percent of known craters in that region and found about 6,000 new ones.
Previous computer algorithms designed to count craters faster and more accurately than humans weren’t as good at spotting craters in regions the algorithms hadn’t studied in training, says Ari Silburt, an astrophysicist at Penn State.
The new AI, described March 6 at arXiv.org, could also pick out craters in pictures of Mercury.
A moon-trained AI that can apply its knowledge to other solar system bodies could help survey worlds where humans have yet to count many craters, says study coauthor Mohamad Ali-Dib, an astrophysicist at the University of Toronto at Scarborough. Such worlds may include Mars, the asteroid Vesta, the dwarf planet Ceres or the icy moons of Jupiter or Saturn.
An AI that inspected images of lunar terrain (left) found previously discovered craters (right, blue circles), as well as potential new ones (red circles).
