Urban Functional Zone Delineation Aided by Vision Foundation Models

The delineation of urban functional zones is essential for urban planning and management. Most current methods rely on points of interest data, which tend to be concentrated in city centers, resulting in poor performance in suburban areas. Moreover, significant information from remote sensing images of target regions is often overlooked. To address these limitations, we propose a flexible approach that leverages existing points of interest data and semantic segmentation model to create a dataset linking image information to functions. The trained model can classify regions based on varying levels of detail—ranging from general classifications for entire areas to more detailed classifications for individual buildings or lands. In addition, we incorporate a super-resolution model to enhance satellite image resolution, addressing the challenge of poor semantic segmentation due to low image quality. This method improves functional zone delineation by combining points of interest data and imagery information, refining image quality for more accurate mapping and enhancing performance in suburban areas.