Map Generalization Method Supported by Graph Convolutional Networks

Abstract. Map generalization has always been a key research issue in cartography. With the continuous development of the information age, massive amounts of map data are being generated, and how to effectively achieve multi-scale representation of large-volume vector data of various types has become a pressing challenge. Traditional methods of map generalization, which rely heavily on human-specified rules and set thresholds, tend to be complex and inefficient. Furthermore, they are often significantly influenced by the subjective factors of cartographers. To address these challenges, this study introduces graph-based deep learning techniques into the field of map generalization. Tailored generalization strategies were designed for point features, polyline features, and polygon features, enabling this data-driven approach to facilitate map generalization tasks from different perspectives. A comprehensive map generalization framework was developed for various feature types by integrating domain knowledge with data-driven techniques. This framework includes the construction of graph structures for different geographic objects, the extraction of feature vectors, and the design of deep learning network models. Experimental results demonstrate that the proposed method delivers good visual performance while preserving the various characteristics of the original map during the generalization process.