Artificial intelligence (AI) has been used to find thousands of new craters on the Moon that had previously gone undetected.
Described in a paper on arXiv, researchers led by Ari Silburt and Mohamad Ali-Dib at the University of Toronto, Scarborough, used an algorithm to scour 90,000 images of the Moon and look for craters that were wider than 5 kilometers (3 miles).
The results revealed nearly 7,000 craters, which almost doubles known craters of this size on the Moon. And it’s hoped that improving the technique could find hundreds of thousands of smaller craters here and elsewhere.
“Crater counting on the Moon and other bodies is crucial to constrain the dynamical history of the Solar System,” the team wrote in their paper.
“This has traditionally been done by visual inspection of images, thus limiting the scope, efficiency, and/or accuracy of retrieval.
“Our results suggest that deep learning will be a useful tool for rapidly and automatically extracting craters on various Solar System bodies.”
They note that on bodies like the Moon, Mercury, Ceres, and Vesta, the lack of an atmosphere has meant that numerous craters have built up over time. But additional impacts can make them hard to spot.
Normally craters are found by manually looking through images. However, this is tricky for smaller craters a few kilometers or less in size. It can also lead to high errors, with differences in total crater numbers as high as 40 percent.
People have developed crater detection algorithms (CDAs) before to sift through data, but they’re not always that accurate. So this team created a convolutional neural network (CNN), a more advanced algorithm that can sort of teach itself to look for craters. CDAs need to be trained on existing images, but the CNN could learn to study regions it hadn’t seen before.
The results were impressive. Looking at images showing elevation maps of a third of the Moon for just a few hours, the CNN found 6,883 new craters. It did make a few errors, including incorrectly identifying some craters and failing to spot some larger ones, but the early signs are promising.
“We anticipate that the uncharted territory of systematic small-size craters identification will provide important new information about the size distribution of Lunar impactors and the formation history of the Moon,” the team concluded.
And it’s hoped that future advancements of the technique may allow for the identification of craters on other worlds where an elevation map is available.
[H/T: New Scientist]
1