基于图卷积神经网络的建筑物合并方法OACSTPCD

Building aggregation is a complex decision-making problem in map generalization,where rule-based aggregation methods attempt to address the questions of"which to aggregate"and"how to aggregate"through a set of cognitive parameters and constraints.However,due to the complexity of regional environments,the diversity of operator selections and the variability of parameter adjustments,the results of rule-based aggregation are often unsatisfactory.With the support of artificial intelligence technology,introducing deep learning methods offers a way to overcome the limitations of rule-based approaches from a data-driven perspective.We presented a method for building grouping and ag-gregation based on graph convolutional neural network.We utilized constrained Delaunay triangulation and the minimum spanning tree to con-struct graph structure,extracting building node features and distance features from the graph.Then,we took the node cosine similarity obtained through graph convolution layers and dot product operations,concatenated with distance features as edge characteristics.Finally,we achieved a binary classification result through a fully connected layer.Through comparative experimental analysis,the ARI index of this method is higher than that of existing methods(from 0.674 to 0.800).