NASA and IBM launch open-source lunar research model for ice and crater detection

NASA and IBM have jointly released an open-source artificial intelligence model designed for lunar research, capable of identifying ice regions and craters on the Moon. The model, available on Hugging Face, was developed through a collaboration aimed at supporting scientific research for future lunar missions. IBM’s Juan Bernabe-Moreno highlighted that the model excels at detecting potential ice areas, achieving a 23% reduction in error compared to Microsoft’s SwinV2-B model. It also outperformed SwinV2-B by 19% in crater recognition using half the training data. Bernabe-Moreno explained the challenges of training the model, including the stark contrast between light and shadow on the Moon’s surface and the difficulty of using traditional training methods due to the similarity of crater appearances. The team solved this by dividing the Moon into sectors for training and testing. The model was tested following a SpaceX rocket impact on the Moon on August 5, successfully identifying the new crater formed by the collision. The release includes the first open-source dataset for lunar research, combining data from NASA’s LRO and GRAIL missions, as well as Japan’s SELENE probe. Bernabe-Moreno emphasized that this dataset, with over 2 million data points, represents a major scientific achievement, providing a foundation for future AI models.