Enhancing semantic information of vector road networks using tile maps

Semantic information is a critical component of spatial data; however, it is often incomplete, limiting the usability of vector road networks. Tile maps are freely accessible, information-rich raster data containing complete road annotations. However, optical character recognition (OCR) applied to tile maps often produces fragmented text boxes and detects non-road text owing to resolution limitations and varying text orientations. To address these challenges, a method for enhancing the semantic information of vector road networks using tile maps is proposed in this study. The vector road network and tile maps are co-registered and segmented. An improved text extraction and merging strategy based on PaddleOCR is developed using geometric constraints to reconstruct fragmented road-name text. Furthermore, a geometry-constrained text classification model, termed the RoadSense Classifier, is introduced to filter non-road text and retain accurate road names. Matching the extracted road text to its corresponding road segments enriches the semantic information. Experiments using tile map data from the Third Ring Road area of Zhengzhou City and the main urban area of Xinxiang show that the proposed method achieves a precision of 98.86% and 97.64%, recall of 85.07% and 83.98%, and F1-score of 91.45% and 89.32%, respectively, improving vector road attribute completeness.