首页 > , Vol. , Issue () : -
Accurate extraction of urban functional zones (UFZs) and a comprehensive understanding of their spatial distribution play an important role in urban planning and management. This paper proposes a UFZ classification method combining object unit and Vision Transformer to address the problem. Firstly, this method utilizes the over-segmented object generated from the multi-scale segmentation method as analysis units to avoid that there are multiple kinds of UFZs in an object. Then, considering that current methods always focus on the inherent analysis of objects and ignore spatial relationships among them, Transformer is employed for spatial relationship modeling between objects, in which the geographic attributes of objects act as position embedding. In this way, the inherent features of a single analysis unit and inter-spatial features among objects are both taken into account for UFZ classification. The experimental results show that, compared with the results of the existing methods, the over-segmented objects can improve the boundary accuracy, avoiding the jagged boundaries resulting from grid units and the multi-UFZs in a single unit resulting from road-block units. Besides, the accuracy of UFZ classification increases by 13.9% to the method employing objects as analysis units and ignoring their spatial relationships.