面向属性空间分布特征的空间聚类

发布时间：2018-07-06 16:17

本文选题：空间聚类 + Delaunay三角网　；参考：《遥感学报》2017年06期

【摘要】：空间聚类应当同时满足空间位置邻近和属性相似,在此背景下,为满足空间邻近实体之间趋势性和不均匀性的属性聚类需求,提出一种基于图论和信息熵的空间聚类算法。该算法主要是在Delaunay三角网空间位置聚类基础上,通过引入信息熵,采用多元相似性度量方法以解决二元关系在属性聚类中的缺陷,同时基于"等概率最大熵"原则提出了一种局部参数度量方法,用于表达邻近目标间属性分布的局部变化信息。将本文方法与多约束聚类方法和DDBSC聚类方法进行对比分析,结果表明:(1)在属性空间分布不均的情况下,本文方法的聚类精度要高于多约束方法和DDBSC方法,尤其是当属性空间分布不均程度不断扩大时,DDBSC和多约束算法会将空间簇内的实体误判为噪声;(2)在对异常值的敏感性问题上,3类方法都能识别出异常值的位置,但DDBSC和多约束算法对异常值具有一定的敏感性,聚类结果会掩盖属性分布的趋势性,本文方法受异常值影响很小。通过模拟实验和实际算例可以发现,在保证空间邻近的基础上本文方法具有如下优势:第一,能反映实体属性在空间分布中的趋势性特征;第二,能满足属性空间分布不均匀;第三,对异常值具有良好的稳健性。
[Abstract]:Spatial clustering should satisfy both spatial location proximity and attribute similarity. In order to meet the demand of attribute clustering between spatial adjacent entities, a new spatial clustering algorithm based on graph theory and information entropy is proposed. The algorithm is mainly based on Delaunay triangulation spatial location clustering, by introducing information entropy, adopting multivariate similarity measure method to solve the defect of binary relation in attribute clustering. At the same time, based on the principle of "equal probability maximum entropy", a local parameter measurement method is proposed to express the local variation information of attribute distribution between adjacent objects. The results show that: (1) the clustering accuracy of this method is higher than that of multi-constraint method and DDBSC method under the condition of uneven distribution of attributes in space, and the clustering accuracy of this method is higher than that of multi-constraint method and DDBSC method, the results show that: (1) the clustering accuracy of this method is higher than that of multi-constraint method and DDBSC method. Especially, DDBSC and multi-constraint algorithms will misjudge the entities in the spatial cluster as noise when the degree of spatial distribution is increasing. (2) on the sensitivity of outliers, all three methods can identify the position of outliers. However, DDBSC and multi-constraint algorithms are sensitive to outliers, and the clustering results cover up the tendency of attribute distribution. The method in this paper has little effect on the outliers. Through simulation experiments and practical examples, it can be found that the method has the following advantages: firstly, it can reflect the trend characteristics of entity attributes in spatial distribution, second, it can satisfy the non-uniform spatial distribution of attributes, and the method has the following advantages: firstly, it can reflect the trend of the physical attributes in the spatial distribution, second, it can satisfy the non-uniform spatial distribution of the attributes. Thirdly, it has good robustness to outliers.
【作者单位】：南京师范大学虚拟地理环境教育部重点实验室;江苏省地理信息资源开发与利用协同创新中心;
【基金】：国家自然科学基金(编号:41671392) 公安部科技强警基础工作专项项目(编号:2015GABJC39)~~
【分类号】：P208

【相似文献】