Digital Sand Table Model Construction Based on Improved Transformer Model: Multi-Scale Feature Fusion of Terrain and Line Data

A hybrid map–transformer framework, GTRM-Net, is introduced for terrain-aware path planning within intelligent transportation and urban governance systems. The framework integrates multi-scale digital elevation models (DEMs) and road network structures through a bimodal encoder, where a graph neural network captures topological and elevation attributes, and a convolutional neural network extracts terrain features from topographic imagery. These representations are fused and decoded via a Transformer architecture to generate optimal path sequences, with elevation regularization enhancing terrain adaptability. Experiments on SyntheticGrid and OSMnx datasets demonstrate substantial improvements in path accuracy, efficiency, and elevation management, highlighting GTRM-Net’s potential for intelligent transportation, tactical logistics, and emergency routing applications.