DR-QGNN: A Quantum Graph Convolutional Neural Network with Data Re-uploading for Road Network Selection

The intelligent selection of road networks is crucial in cartographic generalization, but due to the high complexity of road networks, existing deep learning methods have not effectively solved this problem. Given the advantages of quantum computing in certain tasks, we propose a Data Re-uploading Quantum Graph Convolutional Neural Network (DR QGNN) for the intelligent selection of road networks. By repeatedly re-encoding input features within the quantum graph convolution layer, DR-QGNN enhances its expressive power. Unlike traditional QGNNs that require a one-to-one correspondence between qubits and graph nodes, DR-QGNN encodes node sequentially into a quantum circuit with fewer qubits. We introduce external indicators reflecting road context, Residential Area Oriented Transport Capacity (RATC) and Vertical Traversability Difficulty (VTD), to capture both road functionality and terrain complexity. The hybrid architecture integrates the classical concept of graph aggregation with quantum circuit evolution. Validated on Hong Kong road data at multiple map scales, DR-QGNN outperforms conventional methods across five key metrics: accuracy, recall, F1, CLC, and connectivity. Ablation studies and cross-region evaluation further demonstrate the efficacy of DR-QGNN and contextual indicators, highlighting DR-QGNN’s broad applicability and offering a new quantum graph neural network approach for geospatial vector data modeling.