A Road Networks Similarity Calculation Method Based on Graph Convolutional Autoencoder

Objectives: The calculation of spatial similarity relationship of road networks has wide applications in the fields of spatial data matching and querying, spatial data multi-scale expression and evaluation, and Quality evaluation of cartographic generalization. However, the existing road networks similarity calculation methods have the problems of insufficient utilization of spatial information and too subjective setting of the weights of feature factors. The graph convolutional autocoder adopts a self-supervised approach to training, with the goal of reconstructing the original input graph as much as possible and realizing end-to-end learning, which is capable of overcoming the limitations of the traditional approach. Therefore, a road network similarity calculation model based on graph convolutional autoencoder was proposed. Methods: The model firstly constructs the dual graph of the road networks, and based on the principle of the relationship between the whole and the part of the structure, assigns the spatial feature information of the road network to the nodes of the dual graph from three aspects of the global, local and connectivity characteristics to get the quantitated expression of the structure of the road networks graph; secondly, the graph convolutional autocoder is used to aggregate and update the node feature information and the structural information of the road networks graph, to form the depth of road networks cognition, and get the coded expression of the road networks node information; finally, use the average pooling operation to map the complex high-dimensional feature space to the easy-to-measure low-dimensional feature space to get a set of feature vectors, and use the cosine similarity to calculate its similarity. Results: The experimental results show that the similarity calculated by this method has high sensitivity and maintains high consistency with the actual road changes, which is more in line with human cognition. Conclusions: The study shows that the model has a high degree of automation, the calculation results are reasonable and intuitive, which greatly reduces the design of manual features and rules, and can effectively realize the calculation of road network similarity.