高分辨率多光谱遥感影像细粒度建筑物特征数据集

高分辨率多光谱遥感影像细粒度建筑物特征数据集选择覆盖了总共3668平方千米的中国各个灾害带的21个区县建筑物集中区域为研究区，从光谱、纹理、边缘、指数四个角度，生成了17种特征分量。
高分辨率多光谱遥感影像细粒度建筑物特征集中共包含超过26万个建筑物实例，大规模的数据保证了较高的类内差异，包括尺寸、形状、颜色、角度、背景等差异，为后续泛化性模型的建立提供了数据支撑。此外，高分辨率多光谱遥感影像细粒度建筑物特征集特有地将建筑物按照结构类型分为钢及钢筋混凝土结构、砌体结构以及砖石和其他结构三种。不同结构类型的建筑物抵御灾害的能力以及可使用时间区别显著，细粒度的设计使得遥感提取建筑物任务将发挥更大的作用，尤其是灾害领域的灾前损失预测和灾后损失评估。

This high-resolution multispectral remote sensing imagery fine-grained building feature dataset selects 21 concentrated building districts and counties across various disaster zones in China, covering a total area of 3668 square kilometers as its study area. Seventeen feature components are derived from four dimensions: spectral, texture, edge, and index.
The dataset contains over 260,000 building instances. The large-scale nature of the dataset ensures significant intra-class variations, including differences in size, shape, color, orientation, background and other aspects, providing robust data support for the development of subsequent generalizable models.
Furthermore, this dataset uniquely categorizes buildings into three structural types: steel and reinforced concrete structures, masonry structures, as well as brick, stone and other miscellaneous structures. Buildings of different structural types exhibit notably distinct disaster resistance capabilities and service lifespans. The fine-grained design enables remote sensing-based building extraction tasks to play a more pivotal role, particularly in pre-disaster loss prediction and post-disaster loss assessment within the disaster management domain.