Crater counting on the Moon is an important part of understanding the dynamical history of the Solar System. This is conventionally undertaken through detailed, and painstaking, visual inspection of images by scientists. While accurate to a degree, this limits the scope, efficiency, and accuracy of the information collated.
In a new study, researchers assessed the potential of using convolutional neural networks to determine the positions and sizes of craters on the Moon. The information was taken from a Lunar digital elevation map.
The machine learning platform discovered near double the total number of crater detections that the human centric approach had achieved. Many of these carters were small, suggesting that a machine can discern smaller structures than a person. In all, 361 new craters were discovered.
Furthermore, the artificial intelligence has unveiled thousands of new pockmarks across on the Moon’s surface. This program also has the functionality to catalog impact scars from collisions from space objects. Such information should improve scientists’ understanding of how different objects roamed the Solar System in the past.
Furthermore, assessing Moon related impact damage provides valuable information about the history of the Earth. For instance, a barrage of rocks hitting the solar system 3.9 billion years ago, as seen from the lunar surface, could also have dramatically reshaped Earth’s geology and atmosphere.
The technology was developed by Ari Silburt, an astrophysicist at Penn State, and Mohamad Ali-Dib, an astrophysicist at the University of Toronto at Scarborough.
The researchers expect that this form of deep learning can become a reliable method for rapidly and automatically extracting craters on other Solar System bodies. This should extend to Mars, the asteroid Vesta, the dwarf planet Ceres or the icy moons of Jupiter or Saturn. In addition, the researchers have made the code freely available to interested parties.
The research has been published as a white paper by Cornell University Library. The paper is titled “Lunar Crater Identification via Deep Learning.”
