A point cluster simplification approach of graph convolutional neural network for map generalization
Map generalization is a complex decision-making process with multiple factors,and the optimal selection of generali-zation operators for different scenarios is routinely implemented through a rule-based approach.These map generalization rules need to be"patched"to take into account the influence of different special conditions,resulting in an increasingly complex sys-tem of map generalization rules that lose their universality.The data-driven simplification scheme with artificial intelligence technology provide a new way of thinking about map generalization under special rules by extracting the generalization rules im-plied in typical cases through machine learning and migrating them to new data scenarios.In this paper,deep learning tech-niques are introduced and a strategy combining domain knowledge and data-driven is used to propose an automatic generaliza-tion method for point clusters based on graph convolutional neural networks.This method obtains the knowledge of map gener-alization in different data scenarios through sample training and deep learning,while incorporating established rules for guid-ance,which can move more effectively toward the goal of artificial map generalization results.Firstly,a Delaunay triangulation network is constructed to establish spatial neighbourhood relationships between points and to calculate the feature information of each point based on domain knowledge such as geospatial contextual associations and spatial heterogeneity to construct the feature vectors of the point cluster.Secondly,the topological adaptive graph convolutional neural network is introduced to con-struct an automatic generalization network model of the point cluster data.The experiments show that the algorithm can main-tain the features of the original point cluster in both the local area and the overall map,which is reflected in the relative quantity maintenance,contextual feature inheritance and attribute feature consistency.
万方数据
维普
