Similarity Judgment Of Linear Arrangement Of Buildings Based On Graph Convolution Neural Network

As an important part of spatial relations,the research of spatial similarity has always been the focus in the field of geographic information science because of its indispensable position in spatial relations and its rich theoretical and application value.Urban building is a kind of spatial object widely existing in geospatial space,and it is an important component of geospatial data.The spatial distribution pattern of urban building group is also a research object with rich practical significance.As the most common pattern of spatial distribution of buildings,linear arrangement of urban buildings widely exists in geographical space.Its similarity calculation is of great value to cartographic generalization and map matching.The similarity judgment of spatial distribution pattern needs the support of quantitative calculation.The traditional spatial similarity calculation methods usually need to set the weight of measurement factors.The calculation of the similarity degree of spatial objects is closely related to the setting of weights,but the reasonable setting of weights has always been a problem.In recent years,with the development of deep learning,many scholars use deep learning methods to analyze and process geospatial data,GCNN)solves the problem that Geospatial Vector data is not suitable for deep learning method,and deep learning method can reduce the influence of human factors on similarity measurement.Therefore,this paper uses graph convolution neural network method to explore and study the similarity judgment of linear arrangement of buildings(1)The division of building spatial distribution pattern.This paper introduces and analyzes the existing research on the spatial distribution pattern of buildings.From the perspective of residential activity area,buildings are divided into urban central area buildings,urban suburban buildings and rural area buildings.According to the demand of the similarity calculation of the linear arrangement of buildings,the linear distribution pattern of buildings is divided.(2)A description model of linear arrangement of Buildings Considering the principle of visual cognition.This paper measures the linear arrangement of buildings from two aspects:from the perspective of individual buildings,it defines four measurement indexes of the size,shape,direction and density of buildings;from the perspective of building groups,it selects five degrees for the linear arrangement of buildings: the direction,length,uniform degree of size,the deviation degree of buildings and the consistency degree of buildings and paths Quantity index to describe the structural characteristics.(3)Recognition and extraction of linear arrangement of buildings.Delaunay triangulation is used to analyze the neighborhood relationship of building groups,and MST method is used for rough clustering of building groups to realize the preliminary extraction of linear arrangement of buildings.Finally,through the consistency of building and path direction as constraint parameters,the rough clustering results are refined clustering,and the linear arrangement of buildings is obtained.(4)GCNN is used to judge the similarity of linear arrangement of buildings.Combined with graph convolution operation and traditional convolution network model,the GCNN model of building linear distribution pattern similarity judgment is established.Through the analysis and calculation of the parameters describing the model,the characteristic matrix and adjacent matrix of GCNN model are obtained,and the classification label is obtained according to the volunteers' voting,which constitutes the experimental data set of graph convolution neural network model.The GCNN model is used to judge the similarity of linear arrangement of buildings.