MRGC-MCINet: multi-relation graph convolutional neural network with Monte Carlo inference for automatic road network selection

Automatic road network selection is a pivotal enabling method for the efficient acquisition of multi-scale road network data. This paper proposes MRGC-MCINet, an end-to-end road network selection model that integrates multi- relation graph neural networks and Monte Carlo inference. First, the vector linear road network is modelled as a graph structure, and edge features are integrated with node features to construct the road network feature matrix. Second, a multi-relation graph convolutional neural network is designed to learn multi-dimensional features and realise road selection via a binary classification task. Finally, Monte Carlo model is incorporated for multi-path inference to estimate the uncertainty of the road network selection model and improve the robustness of road selection decision. The experimental results demonstrated that MRGC-MCINet outperforms the classic stroke-mesh hybrid selection method and the standard GCNN-supported selection method, validating its effectiveness and feasibility in road network selection tasks.