尺度驱动的空间聚类理论

发布时间：2018-03-01 00:19

本文关键词： 空间聚类尺度自然法则视觉认知假设检验　出处：《测绘学报》2017年10期 　论文类型：期刊论文

【摘要】：空间聚类是探索性空间数据分析的有力手段,不仅可以直接用于发现地理现象的分布格局与分布特征,亦可以为其他空间数据分析任务提供重要的预处理步骤。空间聚类有望成为大数据认知的突破口。空间聚类研究虽然已经引起了广泛关注,但是依然面临两大最根本的困境:"无中生有"和"无从理解"。"无中生有"指的是:绝大多数方法,即使针对不包含聚类结构的数据集,仍然会发现聚类;"无从理解"指的是:即使同一种聚类方法,采用不同的聚类参数就会获得千变万化的聚类结果,而这些结果的含义不明确。造成上述困境的根本原因在于:尺度没有在聚类模型中被当作重要参数而恰当地体现。为此,笔者受到人类视觉多尺度认知原理的启发,根据多尺度表达的"自然法则",建立了一套尺度驱动的空间聚类理论。首先将尺度定量化建模为聚类模型的参数,然后将空间聚类的尺度依赖性建模为一种假设检验问题,最后通过控制尺度参数以自动获得统计显著的多尺度聚类结果。在该理论指导下,可以构建适用不同应用需求的多尺度空间聚类模型,一方面降低了空间聚类过程中的主观性,另一方面有利于对空间聚类模式进行全面而深入的分析。
[Abstract]:Spatial clustering is a powerful tool for exploratory spatial data analysis. It can not only be directly used to discover the distribution pattern and characteristics of geographical phenomena. It can also provide important preprocessing steps for other spatial data analysis tasks. Spatial clustering is expected to become a breakthrough in big data's cognition. Although spatial clustering research has attracted extensive attention, But there are still two fundamental dilemmas: "created by nothing" and "incomprehensible." "myth" refers to the vast majority of methods, even for data sets that do not contain clustering structures, Clusters will still be discovered; "incomprehensible" means that even with the same clustering method, the use of different clustering parameters results in a variety of clustering results. But the meaning of these results is not clear. The root cause of the above dilemma lies in the fact that the scale is not properly reflected as an important parameter in the cluster model. Therefore, the author is inspired by the human visual multi-scale cognitive principle. According to the "natural law" of multi-scale expression, a set of scale-driven spatial clustering theory is established. Firstly, the scale quantification is modeled as the parameter of the clustering model, and then the scale dependence of spatial clustering is modeled as a hypothesis testing problem. Finally, the statistical significant multi-scale clustering results are obtained automatically by controlling the scale parameters. Under the guidance of the theory, a multi-scale spatial clustering model suitable for different application requirements can be constructed. On the one hand, the subjectivity in the spatial clustering process is reduced. On the other hand, it is beneficial to the comprehensive and in-depth analysis of spatial clustering model.
【作者单位】：香港理工大学土地测量与地理资讯学系;西南交通大学高铁运营安全空间信息技术国家地方联合实验室;中南大学地理信息系;
【基金】：国家自然科学基金(41601410;41471383) 湖南省自然科学基金(2017JJ3379)~~
【分类号】：P208

【相似文献】