Chinese scientists are deeply integrating artificial intelligence into the field of lunar science research. During the 2025 China International Big Data Industry Expo, the Institute of Geochemistry of the Chinese Academy of Sciences released the "Multimodal Professional Large Model V2.0 of Lunar Science" on August 29. The construction of this embedded professional large model endows the "Digital Moon" cloud platform with an efficient "intelligent brain" for operation. Human research on the geological evolution of the moon relies on the analysis of geological structures such as impact craters. The number, size, depth, shape, and other characteristics of impact craters are the key basis for analyzing the history and effects of lunar impacts. Liu Jianzhong, a researcher at the Institute of Geochemistry of the Chinese Academy of Sciences, said that at present, more than 1 million impact craters with a diameter of more than 1 km have been identified on the moon, and the number of impact craters with a diameter of less than 1 km cannot be accurately counted up to now. If relying solely on manual identification, it is almost impossible to complete the investigation of all lunar impact craters. The application of specialized lunar models has greatly improved research efficiency. Researchers only need to input images of lunar impact craters and related questions, and the specialized models can answer questions about the shape, size, and formation age of the impact crater, and provide complete textual descriptions simultaneously. At the press conference, Liu Jianzhong introduced that in the research process of professional large-scale models, a comprehensive and systematic multimodal data annotation standard has been established, and a labeled instruction dataset containing over 8700 lunar impact craters and 7272 other lunar structures has been constructed. Breakthrough progress has also been made in model training, "said Liu Jianzhong." The accuracy of the large model in the task of classifying impact crater ages and subtypes has reached 88%, and the accuracy in the task of automatically identifying lunar structures has reached 93%. "According to the project plan, the" Digital Moon "cloud platform will be completed in 2027 and open for global sharing. (New Society)