(ORDO NEWS) — An international team of researchers used machine learning algorithms to count and map over 100,000 lunar craters.
Previous work on the identification and localization of craters on the Moon has shown that this process is very long – usually it was done manually, by studying photographs and then compiling maps using the information received.
In a new study, scientists have found a way to dramatically speed up this process by training a computer to identify craters and then count them.
Teaching a computer to recognize craters is challenging because craters can take many forms. Not all craters are regular rings, while all craters are of different ages, and therefore, in the most ancient craters, the defining characteristics were “washed out” under the influence of erosion processes.
Scientists wanted to map the location of all the craters on the lunar surface and date each crater to provide a powerful tool for studying the history of our solar system.
This new approach involved “training” a machine learning algorithm to recognize the underlying structure of the crater. Then the algorithm was “trained” to recognize craters in a broader context, based on the analysis of data obtained using the Chinese lunar orbiters Chang’e-1 and Chang’e-2.
After training the system was finally completed, the researchers applied it to the analysis of data collected using the Chang’e-5 lander, which was part of the Chinese mission to return soil samples from the lunar surface. A machine learning algorithm used this data to identify and count craters at mid and low lunar latitudes. This new system counted a total of 109,956 craters – far more than has ever been counted for the Moon.
The research is published in the journal Nature Communications; lead author Chen Yang.
Contact us: firstname.lastname@example.org