拉普拉斯特征映射的时空数据划分方法  

Spatio-temporal data partitioning method based on Laplacian eigenmaps

作　　者：夏慧琼[1,2] 林丽群 XIA Huiqiong;LIN Liqun(Faculty of Resources and Environmental Science, Hubei University, Wuhan 430062, China;Key Laboratory of Regional Development and Environmental Response in Hubei Province, Wuhan 430062, China)

机构地区：[1]湖北大学资源环境学院,武汉430062 [2]区域开发与环境响应湖北省重点实验室,武汉430062

出　　处：《测绘科学》2018年第6期32-38,共7页Science of Surveying and Mapping

基　　金：湖北省自然科学基金项目(2014CFB560);湖北省教育厅基金项目(Q20141003)

摘　　要：针对传统的时空数据划分方法造成地理对象时空关系的割裂、时空信息维护困难等问题,该文提出一种基于拉普拉斯特征映射(LE)的时空划分方法。选取空间位置、时间、属性及空间拓扑关系邻近性等主要时空描述信息作为表达时空关系邻近性的因素,分析各因素对时空查询的影响程度,采用多因素结合方法构建无向邻域图并建立Laplacian矩阵,通过对矩阵特征分解将时空信息映射至二维谱空间,并在此基础上进行时空数据的划分。实验结果表明,该方法能够较好地保持地理对象时空关系的完整性,能够获取数据在低维流形空间中的结构特征,使得时空信息的组织和管理变得更加容易。Aiming at the problem that traditional spatio-temporal data partitioning methods resulted in seperated relationships between space and time of geographical objects and difficult spatio-temporal information maintenance,according to the problems,a new method on spatio-temporal data partitioning based on Laplacian eigenmaps（LE）was proposed.Several spatial and temporal descriptive factors,such as space position,time,attribute and spatial topological relations were chosen to indicate spatio-temporal neighbor relationship,the impact and contribution of each factor on spatio-temporal query were analyzed,then the undirected neighbor graph and Laplacian matrix were built based on a multi-factor fusion method,the spatio-temporal information was mapped to two-dimensional spectral space by performing eigen-decomposition of the Laplacian matrix,and division of spatio-temporal data was implemented based on the characteristic of mapped data.The experimental results showed that this method could better maintain the integrity of spatio-temporal relations between geographic objects,and the structural features of the data in low-dimension manifold space could be obtained,which made it easier to organize and manage spatio-temporal information.

关 键 词：时空查询 拉普拉斯特征映射 时空划分 时空邻近 

分 类 号：P208[天文地球—地图制图学与地理信息工程]