PUMA：正样本-无标签学习实现标签高效的地理空间基础模型适配
原标题：Towards Label-Efficient Geospatial Foundation Model Adaptation via Positive-Unlabeled Learning for Building Damage Mapping
AI 摘要
该论文提出PUMA框架，将正样本-无标签（PU）学习与参数高效微调（PEFT）结合，用于地理空间基础模型（GFM）的建筑损毁制图适配，仅需少量已确认损毁建筑的标注。PUMA通过非对比弱监督表示精炼和基于隐状态的标签消歧两个相互增强的组件，缓解弱监督噪声对表示学习的影响，并从广义期望最大化（GEM）视角给出理论解释。在全球灾害基准数据集上的实验验证了其有效性与鲁棒性，2025年洛杉矶山火的真实应用进一步显示其在极少监督下准确制图的潜力。
正文节选
Towards Label-Efficient Geospatial Foundation Model Adaptation via Positive-Unlabeled Learning for Building Damage Mapping Abstract Geospatial foundation models (GFMs) are reshaping remote sensing by shifting the field from task-specific modeling toward general-purpose geospatial intelligence. Mapping the damage to buildings stands to benefit substantially from this shift, as GFMs pretrained on large-scale Earth observation data provide generalizable geospatial representations that can facilitat