Comparison of Deep Learning Models for Crater Detection Using OHRC Dataset of Chandrayaan-2

The Orbital High-Resolution Camera (OHRC), with a spatial resolution of 25 cm, is one of the most sophisticated optical sensors available for lunar exploration, capturing detailed observations of the lunar surface. This data is crucial for lunar missions to define safe and unsafe regions such as craters, ridges and boulders. The deep learning approach eliminates the need for manual intervention and streamlines the analysis of large lunar datasets. The combination of planetary data and modern processing techniques has significantly improved the accuracy and efficiency of crater detection, enabling detailed and large-scale analyses of planetary surfaces. The study area undertaken for the application of deep learning models for automated crater detection is the landing site of the NASA Apollo 12 (April 1967) and Surveyor 3 (November 1969) missions. Models such as YOLOv8 and Mask R-CNN were applied to detect medium to large-sized craters. The YOLOv8 model result outperformed the Mask R-CNN result by providing higher accuracy and faster execution time. This study contributes valuable insights to the planetary science community by highlighting the robustness of different deep learning models for crater analysis using optical datasets from the Chandrayaan- 2 lunar mission by ISRO, India.