Recognition Method for Road Grid Pattern By Integrating GraphSAGE Model and Gating Mechanism

Abstract

Grid pattern recognition is of great significance for spatial pattern cognition, cartographic generalization, and multi-scale representation. To address the issues that existing methods rarely consider multi-level features and insufficiently utilize the learning and mining capabilities of intelligent models, this paper proposes a method for road grid pattern recognition that combines the GraphSAGE model and the gating mechanism. Specifically, the road network is first converted into mesh polygons, from which the global, local, and neighborhood features of each mesh unit are extracted in sequence. A grid recognition model is then constructed with GraphSAGE as the basic framework, dynamically fusing the multi-granularity features generated by different aggregation functions through the gating mechanism and thereby produce the recognition results. Experimental results on several real-world road network datasets show that the proposed method achieves accuracy, precision, recall, and F1 score of 96.73%–96.99%, 96.57%–96.77%, 97.00%–97.86%, and 96.79%–97.31%, respectively, outperforming both conventional and graph neural network baseline methods. Moreover, benefiting from the multi-level features established in this study, the method can effectively identify special structural patterns such as parallelogram meshes and multi-rectangle meshes at road intersection corners, improving the accuracy and reliability of grid recognition.