Block-Level Matching Recognition Algorithm for OpenStreetMap and Segments From High-Resolution Remote Sensing

With the rapid advancements in remote sensing (RS) and geographic information systems (GIS), their integration has made significant progress. However, RS data faces challenges such as spectral ambiguity and object confusion, while GIS data often lacks sufficient physical detail for mechanistic interpretation. To address these issues, we propose a matching recognition algorithm that combines OpenStreetMap (OSM) data with high-resolution RS imagery. Segmentation methods such as eCognition, the edge-guided image object detection approach, and the segment anything method were utilized to transform RS imagery into vector data. Three filtering criteria—location, area, and spectral characteristics—were applied to match the segmented polygons with OSM features. Optimal polygon pairings were determined based on five matching factors. Controlled experiments employing accuracy and recognition ratio metrics were conducted to evaluate the performance. The results indicate that block-level matching recognition substantially reduces misclassification compared to pixel-level methods, achieving over 60% accuracy across four study areas. The effectiveness of segmentation algorithms influences the results, necessitating the selection of algorithms based on specific scenarios. While different combinations of matching factors do not significantly alter the outcomes, region-specific conditions may require testing various combinations to optimize performance. The collaborative recognition of OSM and high-resolution RS imagery improves object differentiation from both physical and social perspectives, providing a bridge for further analysis and applications in the integrated RS-GIS field.