Intelligent recognition method for urban road grid patterns by fusing mesh and road features

The road grid pattern (RGP) is a common spatial structure in urban road networks, and is important for revealing urban development, population distribution, and functional area delineation. However, recognizing RGPs in vector data is a complex and intelligent task that requires comprehensive consideration of road geometry, semantic features, and their contextual relationships. Previous studies have overlooked the hierarchical nature of the cognitive process, resulting in limited accuracy in recognizing complex grid patterns. For this reason, this study proposes a recognition method that integrates both mesh and roadway features. Initially, from a holistic cognitive perspective, the road mesh is considered as the research object. A model is constructed using topology adaptive graph convolutional networks to obtain preliminary RGP recognition results. Then, the research focus is refined to the road segment level, where the preliminary results are optimized by extracting the geometric and directional features of the road segments. This method significantly outperformed methods based on single research objects across several evaluation metrics, with a performance improvement ranging from 0.79–44.87%. Notably, when dealing with RGPs with complex relationships, this method effectively excludes road segments with irregular internal structures, thereby improving the accuracy and reliability of RGP recognition.