Analysis Of Spatial Distribution Pattern Of Residential Areas Based On Graph Convolution Neural Network

Map spatial distribution pattern is the shape or arrangement of the core geographical elements in the spatial distribution form that can be clearly recognized and recognized by people.Its detection and analysis is one of the basic problems of Geographic Information Science and an important means to mine the hidden spatial knowledge and enhance the existing data.As an area geographic element in Geospatial Vector data,residential area is not only the main carrier of people engaged in production and daily life,but also an important part of Geospatial Vector Database.The spatial distribution pattern of residential areas can not only convey important spatial structure knowledge and spatial cognitive information,but also has a wide range of applications in spatial data mining,spatial reasoning,map generalization and urban planning.The recognition and classification of spatial distribution patterns of residential areas is a innovative key issue,which is mainly reflected in the following three aspects: first,whether in individual form or group distribution structure,the recognition and classification of patterns are closely related to people’s subjective cognition and visual perception;Second,when measuring the pattern,the selection of structural measurement factors and the cognition of residential area pattern classification also have a great impact on the recognition and classification results;Third,as a higher-level abstract concept,spatial pattern also has great uncertainty and fuzziness.Therefore,with the help of the advantages of local feature preservation of deep learning,from the perspective of cartographic generalization knowledge acquisition and spatial data mining,this paper uses deep learning technology to study the recognition and classification method of residential area spatial distribution pattern in large-scale map.The main research work and achievements include:(1)A shape recognition method of individual residential area based on graph convolution neural network is proposed.From the perspective of spatial cognition,with the help of the learning method of depth map convolution,the calculation map is constructed with the contour polygon of area residential elements,and the local morphological features and overall structural features such as the length,direction and spatial position relationship of the contour polygon are extracted as the node attributes of the calculation map.Ten English alphabetic residential areas are used as experimental data sets to build a graph convolution neural network classification model,embed the morphological features of residential area contour boundary into high-dimensional vector,and predict its shape category through classifier.Experimental results show that this method can effectively identify the shape types of residential areas,and the classification results are consistent with human visual cognition.(2)The graph representation of spatial distribution pattern of residential area group based on road network constraints is studied.Based on the Gestalt Visual Gestalt principle and the physical isolation of urban road network and water system,the discrete residential areas are grouped and graphically expressed.When obtaining the proximity relationship between residential groups,in view of the difference in the internal spatial structure between linear arrangement mode and mesh distribution mode,for linear arrangement mode,Delaunay triangulation is constructed with residential contour polygon as the object,while for residential groups with mesh distribution mode,Delaunay triangulation is constructed with the center of residential areas as the node to analyze the proximity relationship of residential areas between groups and generate proximity map,The adjacency matrix of the construction map of residential area group is obtained to realize the graph expression of residential area group.(3)A graph convolution neural network classification model of residential area pattern is proposed.The graph structure of residential area group is constructed based on the neighborhood relationship of residential area group,the size,direction,position and shape of residential area are extracted as the node attributes in the graph,and the graph convolution neural network model for pattern classification of residential area group is established based on graphsage(graph sampling aggregation model).Volunteers annotate the types of residential groups,and select four distribution modes of residential groups: straight line,curve,regular grid and irregular to construct the experimental data set.The experimental results show that the model can identify these four typical spatial distribution modes of residential groups.