Mapping the moon: a deep learning strategy for lunar crater detection from chandrayaan-2 satellite captures

Abstract

The moon has its geological information sustained for billions of years now. With the absence of an atmosphere and no tectonic activities, surface features like craters, rilles, pits, etc. are well preserved. Craters are one of the most predominant features of the lunar surface. Crater counting serves as a crucial methodology, providing key insights into lunar surface evolution, impact history, and other geological processes. With the development of computer vision and image processing, crater counting has switched from the traditional manual counting, which is time-consuming and prone to error, to using deep learning methods with the help of object detection models which has become much faster and precise in just a few years. This project harnesses the power of Convolutional Neural Networks (CNNs) to detect lunar surface craters using satellite images from the Orbiter High-Resolution Camera (OHRC) aboard Chandrayaan 2. The methodology involves image labelling, model training,obtaining weight function during validation, and subsequent testing of the model on unlabelled images. The project leverages YOLOv8’s adept architecture, which is designed to be fast and accurate, making it an excellent choice for a wide range of object detection, instance segmentation, image classification and pose estimation tasks. For the project, more than 750 images taken from the OHRC were manually labelled as crater. Precision and recall were the two metrics used to measure the model’s efficiency. It was observed that as the number of epochs, there were specific epoch numbers where the model’s training performance show higher results as compared to higher or lower epochs. I