Classifying and Categorizing Lunar Craters using Convolutional Neural Networks

Traditionally, lunar crater counting has been done by visual inspection of images of the moon’s surface. This method is time consuming and has poor inter-rater reliability for smaller craters. Automating this process using a Convolutional Neural Network (CNN) greatly improves the speed and reliability at which surface features can be detected and classified. Using available high resolution Digital Elevation Model (DEM) images from the Lunar Reconnaissance Orbiter (LRO), we train a CNN to identify craters and classify them based on the slope of their ejecta blanket. Presently the population of small impactors is not well understood but improved detection of the smallest craters can constrain the size distribution of asteroids in the solar system. Additionally, we intend to search for small craters with novel features that are inconsistent with traditional asteroid impacts to potentially constrain the moon’s interaction history with MACHO dark matter from the Galactic halo.