自组织双重空间聚类算法的城市扩张结构分析应用

Self-organizing Dual Spatial Clustering Algorithm and Its Application in the Analysis of Urban Sprawl Structure

双重空间聚类是能顾及空间连续性和属性相似性的空间数据分析,而常规空间聚类算法难以同时顾及2方面的约束条件。本文采用自组织双重空间聚类算法,对城市扩张结构分析进行了研究。通过改造自组织特征映射的最佳匹配神经元搜索的算法机制,在空间域和属性域进行迭代聚类搜索,实现了自组织双重空间聚类。以武汉市扩张斑块的位置信息和扩张程度指数为输入数据,使用自组织双重空间聚类算法,实现了城市扩张动态结构的识别。自组织双重空间聚类算法使得聚类结果,既在空间域上连续,又在属性域上相近,算法过程具有自组织性,减少了人为影响。

Dual spatial clustering is an exploratory data analysis that deals with spatial contiguity and attributive similarity. Conventional spatial clustering methods cannot perform effective clustering in spatial and attribute do- mains simultaneously. This study employs SOFM （Self-Organizing Feature Mapping） to solve dual spatial clus- tering problems, and then verify the proposed method in the analysis of urban expansion structure. By modifying the algorithm of best matching neuron searching in SOFM, we manage to perform clustering in both spatial and attribute domains. The algorithm includes two independent self-organizing clustering processes. The first one in- cludes a spatial constraint, and the other one includes an attribute constraint. The final result is generated by merging the corresponding two results that derived separately from the two processes. The analysis of the struc- ture of urban expansion of Wuhan city is used as a case study. We feed the proposed model with the location in- formation and the expansion degree information of newly grown urban patches, and the generated dual clustering results could clearly illustrate the spatial structure of urban expansion. As a conclusion, the self-organizing dual spatial clustering method can generate spatial continuous and attributive similar clusters with little artificial inter- ference.