LunarChips
收藏资源简介:
LunarChips是由剑桥大学等多机构联合构建的月球表面多模态集成数据集,旨在为月球基础模型提供机器学习就绪的训练数据。该数据集整合了来自三颗月球轨道遥感任务的六种仪器观测数据,包含18个输入通道,覆盖南纬70°至北纬70°的月球区域,共计20.16万个0.5°×0.5°的空间网格芯片。数据源自NASA行星数据系统,经过严格的空间配准、归一化处理和带状划分,形成训练集、验证集与测试集。数据集通过自监督学习支持月球表面表征学习,可应用于资源勘探、矿物丰度反演、地质单元分类等下游科学任务,为月球原位资源利用和可持续探测提供关键数据基础。
LunarChips is a multimodal integrated lunar surface dataset jointly constructed by the University of Cambridge and multiple other institutions, aiming to provide machine learning-ready training data for lunar foundation models. This dataset integrates observational data from six instruments aboard three lunar orbital remote sensing missions, encompasses 18 input channels, covers the lunar region spanning 70°S to 70°N, and includes a total of 201,600 0.5°×0.5° spatial grid chips. The data, sourced from the NASA Planetary Data System, has undergone rigorous spatial registration, normalization processing and strip partitioning, and is split into training, validation and test sets. The dataset supports lunar surface representation learning via self-supervised learning, and can be applied to downstream scientific tasks such as resource exploration, mineral abundance inversion, geological unit classification and others, providing a critical data foundation for lunar in-situ resource utilization and sustainable exploration.





