A Progressive Clustering Approach for Buildings Using MST and SOM with Feature Factors

Abstract

To address the challenges in current research on spatial clustering algorithms for buildings in topographic maps—namely, their limited ability to effectively accommodate diverse application scenarios, including dense and regular urban environments, sparsely and irregularly distributed rural areas, and urban villages with complex structures—this paper introduces an innovative progressive clustering algorithm framework. The proposed framework operates in a hierarchical manner, progressing from macro to micro levels, thereby enhancing its adaptability and practical versatility. Specifically, it employs the minimum spanning tree (MST) technique for macro-level clustering analysis. Subsequently, a self-organizing map (SOM) neural network is utilized to perform micro-level clustering, enabling a more refined and detailed classification. Within this framework, the minimum spanning tree effectively captures the macroscopic distribution patterns of the building population. The macroscopic clustering results are then utilized as the initial weight configurations for the SOM neural network. This approach ensures that the overall spatial structural integrity is preserved during the subsequent micro-level clustering process. Moreover, the SOM neural network achieves refined optimization of micro-clustering details by incorporating building feature factors. To validate the effectiveness of the proposed algorithm, this study conducts an empirical analysis and comparative testing using building data from Futian District, Shenzhen City. The results indicate that the proposed algorithm exhibits superior recognition capabilities when applied to complex and variable spatial distribution patterns of buildings. Furthermore, the clustering outcomes align closely with the principles of Gestalt visual perception and outperform the comparison algorithms in overall performance.