Recognition Of Urban Functional Areas Combining The Semantic-Spatial Relationship Of Street View Objects And POI

As an important spatial carrier reflecting urban economic and social functions,urban functional area reflects the spatial characteristics of urban functions,and it is an important embodiment of the spatial competition and development form of modern urban economic activities.The identification of urban functional zone is of great significance to urban construction and fine management.Street view image can provide a wealth of detailed features of buildings.Using street view image to identify urban functional areas has always been one of the hot research issues.The existing urban functional area identification methods for street view images only use the visual features of the whole image,but do not take into account the location and interrelationship of each functional part in the region,which makes it impossible for the existing methods to make full use of the most distinguishable visual elements(such as buildings)that make urban functional area identification.In this context,this paper proposes an urban functional area recognition method based on street view target relation perception,which can capture the local spatial relations and global semantic relations in street view images,make full use of the location and semantic features in the image,and improve the ability of the existing methods to identify urban functional areas.In addition,this paper also proposed a semantic-spatial relationship urban functional area identification method combining Street View image and POI data,based on Google Street View image data and Open Street Map POI data for urban functional area identification,the main research contents are as follows:(1)For the main data used in the identification of urban functional areas,such as street view images,POI,urban road network data,etc.,this paper obtains the POI data and urban road network data of the OSM official website by means of crawler,and cleans the data.,extraction,duplicate checking,spatial positioning,definition projection and coordinate transformation and other related processing,combined with geographic information spatial statistical tools,functional area identification and functional structure analysis were carried out.(2)Aiming at the existing urban functional area classification based on street view images,which only focuses on the overall visual characteristics of the image,this paper proposes a relationship-aware network-based urban functional area classification method.The method in this paper mines global semantic relations in labels through a semantic relation network,and captures local spatial relations in images through a spatial relation network.The feature representation is further improved by aggregating the outputs of the two networks.Rather than propagating visual features directly in the network,relation-aware networks explore higher-level feature information.And the effectiveness of the method is demonstrated through geographic visualization and quantitative analysis.(3)In view of the existing POI identification methods based on POI frequency and POI semantic topic model,the spatial location and adjacent information of POI data are ignored,which can not form a better semantic vector of urban functional areas,and thus affect the classification of urban functional areas.This paper proposes an urban functional area identification method combining street view objective and POI semanticspatial relationship.Combining POI data and street view image data,the urban functional area classification method based on relational perception network is used to obtain the urban functional area types of street view images,considering the universality of urban functions and the recognition and significance of data categories.Kernel density analysis method was used to analyze the POI data,and combined with the street view target and POI semantic-spatial relationship of urban functional area identification model to obtain the functional area classification results and visualization analysis.