面向地图综合的多尺度空间聚类理论与方法

Automatic map generalization plays a key role in multi-scale spatial data representation, modeling, management and updating. Currently, automatic map generalization is still one of the most changeling issues in cartography and Geographical Information Science. The map generalization operation requires maintaining the overall structures and patterns presented within the source map. Thus the identification of structures and patterns from the source map is critical for automatic map generalization. Due to the knowledge acquisition bottleneck of the rule-based spatial pattern reorganization methods, discovery of spatial patterns by using spatial clustering has been a hot topic in cartography and GIS. Although some spatial clustering methods were used to recognize the spatial patterns, the coupling relationship between spatial clustering and spatial patterns reorganization is not well investigated. More importantly, the multi-scale characteristic of the spatial pattern and scale transformation rule in map generalization are not well handled by existing spatial clustering methods. On that account, to recognize map structure in support of automatic map generalization, this project aims to develop a methodology for multi-scale spatial clustering based on the physiological model of the human visual perception. Specially, the project mainly consists of five parts: (i) to investigate the definitions of different kinds of spatial patterns and their measures; (ii) to model the effect of scale on spatial clustering for automatic map generalization; (iii) to construct and implement the multi-scale spatial clustering model; (iv) to model spatial constraints in the multi-scale clustering process; (v) to assess the validity of the multi-scale spatial clustering results. In this project, the association relationship among source scale of the map, the multi-scale spatial clustering results and the target scale of map generalization will be constructed. Thus, the multi-scale spatial clustering results (or spatial patterns) are able to provide useful knowledge for automatic map generalization.

地图空间结构模式识别是实现地图自动综合的一个关键问题。采用空间数据挖掘方法提取地图数据中蕴含的空间结构模式是当前地图综合领域的一个重要研究方向。现有研究大多是对空间数据挖掘中聚类分析的简单移植，没有将地图空间结构模式识别涉及的尺度依赖关系及其在地图综合中的尺度转换规则融合于空间聚类分析模型中。为此，本项目结合地图制图的特点与地图综合的原则，基于人类视觉感知的生理学模型，研究一套面向地图综合的多尺度空间聚类的理论与方法，主要包括：（1）地图空间结构模式的度量特征与定量描述；（2）面向地图综合的空间聚类尺度特征建模；（3）多尺度空间邻域构建；（4）带有约束的多尺度空间聚类；（5）多尺度空间聚类的有效性评价。通过本项目的深入研究，建立空间聚类尺度特征与地图综合尺度特征之间的关联关系，从而为地图综合提供从局部到全局的多尺度空间结构模式或知识，提升地图综合的自动化程度和地图综合的质量。