Abstract: Remote sensing hyperspectral data is a typical high-dimensional data with the character of spatial clustering.In the paper，to index this kind of database，the Pyramid Technology（PT） is improved for the application to the mechanism of iDistance indexing.Based on the two techniques above，an optimal index is proposed，called SP-iDistance（Spherical Pyramid based iDistance）.SP-iDistance is a high-dimensional indexing structure in metric space，and adopts two-level spatial partition.The distance from point object to reference point is combined with the SP number which is used to identify the space orientation information to calculate the one-dimensional value mapping from the hyperspectral data.When processing the spectral similarity query，the distance and space direction searching can be accomplished simultaneously.The experimental results demonstrate that SP-iDistance can more efficiently reduce the distance computation and I/O，and has higher filtering efficiency than PT or iDistance indexing individually.It can be used for the real-time applications on hyperspectral data matching.

Key words: high-dimensional indexing, similarity query, remote sensing spectral library

摘要： 遥感高光谱数据是一种具有空间聚集特性的高维数据。对PT方法进行改进使之与iDistance的索引机制相适应，并融合这两种不同的空间划分策略，提出一种适用于高光谱数据的索引结构。该索引是一种度量空间的高维索引，采用两级空间划分，在处理光谱相似性查询时可同时完成针对距离和空间方位的数据过滤。实验证明该索引可以有效降低I/O和距离计算次数，具有较高的剪枝效率，适用于高光谱数据相似性查询。

关键词: 高维索引, 相似性查询, 遥感光谱数据库