Chinese scientists have discovered more than 109,000 craters on the surface of the Moon, dozens of times more than those recognized in the lower and middle regions of our natural satellite. The achievement, according to the magazine «Nature Communications», has been made possible by a new approach to machine learning.
Craters occupy most of the surface of the Moon. However, the manual and automatic methods of detecting their numbers have resulted in inconsistencies as to the exact total. For example, it is often difficult to detect irregular or weathered craters using traditional methods.
Chen Yang’s team from Jilin University in Changchun set out to identify lunar impact craters using a transfer learning strategy, a machine learning approach in which acquired prior knowledge is used to solve an additional problem.
First, the study authors trained a deep neural network using data from 7,895 previously identified and 1,411 dated craters. Using data from the Chinese Chang’e-1 and Chang’e-2 orbiters, the network was able to identify 109,956 new craters, dozens of times more than previously recognized in the mid- and low-latitude regions of the Moon. Of the craters with a diameter greater than 8 kilometers, the network estimated the age of 18,996 of these. The findings have led to the creation of a new database of lunar craters in the mid- and low-latitude regions of the Moon.
The authors suggest that their approach could be adapted for use with other bodies in the Solar System and could help extract more information than is possible with manual analysis methods.