Research On Clustering Algorithm For Area Vector Building Data

As the core component of topographic map,the mining of group structure and distribution characteristics of buildings is the basis of map generalization,map matching and updating,and also an important task before studying regional spatial allocation and analyzing the development trend of buildings.At present,the mining of building distribution pattern can not be directly obtained from spatial data,but is completed on the basis of human thinking and cognition combined with some existing methods.Cluster analysis,as a common technology for building distribution pattern mining,can find the potential structure and similar features in the data set without any prior knowledge,so it is also widely used in building pattern research.Different from point data,the spatial location information of buildings should be considered in the clustering of opposite buildings,and their morphological characteristics should also be combined to make up for the deficiencies of single factor clustering.Meanwhile,most of the existing clustering studies rely on people’s subjective experience when selecting algorithms,while ignoring the relationship between data characteristics and algorithms.Therefore,the selection of clustering algorithm in cartographic synthesis and other practical applications is often blind,which hinders the improvement of automation degree.Therefore,in order to make the clustering algorithm better applicable to the clustering analysis of buildings,and to mine the relationship between the data of different types of buildings and the algorithm,the main work of this paper is as follows:(1)Supporting the construction of geometric model of building clusteringThis part mainly focuses on the construction and expression of building geometric diagram model.The research contents include the construction of building polygon constraint Delaunay triangulation network,the construction of Voronoi-like diagram to reflect the influence range of buildings and the generation method of minimum spanning tree,etc.,which provides geometric support for building spatial pattern mining in the future.(2)Individual characteristic expression and architectural complex characteristic descriptionIn the study of building spatial model,it is necessary to quantitatively express the cognitive parameters in Gestalt principle,which can be divided into two levels.From the micro level,it is necessary to pay attention to the description of the building’s own attributes,including the shape,size and direction of the building.Extraction of these features can help us better understand the similarities and differences between buildings and provide basic data support for building cluster analysis.On the macro level,it is necessary to consider the spatial distribution and interrelation among building groups,including the density,compactness and homogeneity of building groups,and select the appropriate clustering algorithm according to the group characteristics of buildings.(3)Research on building clustering based on multi-feature factorsIn this paper,four typical clustering algorithms,k-means,DBSCAN,MST and HDBSCAN,are selected from numerous clustering algorithms,and the principles and workflow of each algorithm are introduced in detail in combination with building examples.Meanwhile,on the basis of overcoming the defects of traditional clustering algorithms in processing high-dimensional data and giving full consideration to the morphological characteristics of planar buildings,In this paper,a secondary clustering method integrating multi-feature factors is designed,which can be divided into two steps:(1)Based on the location information of buildings,the above four algorithms are used for preliminary clustering,and the spatial proximity of buildings is taken as a reference to divide buildings with similar distances into a group.(2)PCA algorithm is used to reduce the dimension of other feature factors to two-dimensional spatial coordinates,and further accurate division,focusing on ensuring the consistency of building morphology.(4)Building clustering algorithm selection model based on neural networkData of building blocks of different types,densities and distributions are selected from representative areas such as cities,suburbs,villages and urban villages for experiments.The clustering algorithm most suitable for a certain type of buildings is selected through relevant evaluation indexes,and the connection between architectural features and corresponding clustering algorithm is dug out.The selected clustering algorithm is input into the neural network as label data and architectural complex features for training,so as to establish a clustering algorithm selection model.Finally,the trained model is used to predict and verify the new architectural complex data.