Automatic Crater Identification and perimeter prediction using Deep Learning

Lunar craters are bowl-shaped depressions formed by impacts of asteroids, comets, and meteorites. They vary from microcraters, a few meters in diameter, to the South Pole-Aitken Basin, the largest known crater at approximately 2,500 kilometers. Studying these craters reveals insights into the Moon’s impact history, geology, and volcanic activity. This paper presents the use of advanced deep learning techniques for the automatic identification and perimeter prediction of lunar craters. We utilized data from ISRO’s Chandrayaan-2 mission, specifically TMC-2 images, with a training dataset of 2,000 images at 512x512 pixels. Using a customized YOLOv5 model with a 0.5 threshold, we achieved 98.5% precision and 99.3% recall in crater identification, and additionally estimated the total number and perimeter of craters in each image.