"Lunar Science Multimodal Professional Model V2.0" Released

"Lunar Science Multimodal Professional Model V2.0" Released

Published:September 2, 2025
Reading Time:2 min read

Want to read in a language you're more familiar with? Please click the pink icon below for one-click AI translation

On August 29, the Institute of Geochemistry, Chinese Academy of Sciences, officially unveiled the "Lunar Science Multimodal Professional Model V2.0" at the 2025...

On August 29, the Institute of Geochemistry, Chinese Academy of Sciences, officially unveiled the "Lunar Science Multimodal Professional Model V2.0" at the 2025 China International Big Data Industry Expo. This embedded professional model equips the "Digital Moon" cloud platform with an efficient "intelligent brain."

Lunar geological evolution studies heavily rely on analyzing geological structures like impact craters. The number, size, depth, and morphology of these craters are critical for understanding the Moon's impact history and effects.

Researcher Liu Jianzhong from the institute revealed that over one million impact craters with diameters exceeding 1 kilometer have been identified on the Moon, but accurately counting those smaller than 1 kilometer remains challenging. Liu noted that manually identifying all impact craters is "almost an impossible task."

The Lunar Professional Model significantly enhances the efficiency of lunar geological research. Scientists can upload impact crater images and pose related questions, and the model will provide answers about the crater's morphology, size, formation age, and more, accompanied by a comprehensive textual description.

During the launch event, Liu Jianzhong shared that the team established a comprehensive multimodal data annotation standard during the model's development, creating a labeled instruction dataset containing over 8,700 lunar impact craters and 7,272 other lunar structures. "The model training has seen breakthroughs," he said, adding that "the accuracy for crater age classification and subclass division reaches 88%, while automatic identification of lunar structures achieves a 93% accuracy rate."

According to the plan, the "Digital Moon" cloud platform is set to be completed by 2027 and will be openly shared with the global community.