Pattern Recognition in Urban Maps Based on Graph Structures

Map groups exhibit distinct spatial distribution characteristics, making their pattern recognition crucial for map generalization, map matching, geographic dataset construction, and urban planning/analysis. Current pattern recognition methods for map groups primarily fall into two categories: machine learning-based approaches and traditional methods. While both have achieved certain recognition outcomes, they suffer from four key limitations: (1) insufficient algorithmic interpretability; (2) limited model generalizability; (3) restricted pattern diversity in recognition; (4) inability of existing methods (including deep learning and traditional algorithms) to achieve multi-pattern recognition across heterogeneous map group types (e.g., building groups vs. road networks) using a single framework. To address these limitations, this study proposes a graph structure-based multi-pattern recognition algorithm for map groups. The algorithm integrates the quantitative advantages of directional entropy in characterizing spatial distribution patterns with the discriminative power of node degree in analyzing edge-node geometric models. Experimental validation utilized building and road network data from multiple cities, constructing a dataset of 600 samples divided into two subsets: Sample Set 1 (for parameter threshold calibration and rule generation) and Sample Set 2 (for algorithm performance validation and transferability testing). The results demonstrate a classification accuracy of 97% for the proposed algorithm, effectively distinguishing four building group patterns (linear, curved, grid, irregular) and two road network patterns (grid, irregular). This work establishes a novel methodological framework for multi-scale spatial pattern analysis in map generalization and urban planning.