Mixed Urban Functional Region Recognition By Mining Spatial Semantics From Trajectory And Points Of Interest Data

With the development of urbanization in China,the rapid expansion of urban space has produced a series of urban problems.The understanding and exploration of the urban functional structure plays an important role in improving the city and urban planning,and the identification of urban mixed functional areas rapidly and accurately is helpful to the scientific construction and spatial optimization of the city,which promotes the construction of a sustainable city.Recently,data mining technology and urban big data have developed rapidly.Compared to traditional social survey and remote sensing data,the trajectory and point of interest in the city contain massive information about urban land use and spatial distribution and connection.Based on emerging natural language processing tools,these information can be mined effectively to understand urban land use and its degree of mixing more accurately in real time.Therefore,using the mobile phone signaling trajectory and point of interest data in Beijing,this paper quantified and analyzed the city's urban function distributions and the degree of mixing.The framework of geographic semantic feature extraction,mixed urban functional area identification,result verification,sensitivity analysis and result evaluation was built.The main research work of this paper is as follows:First,data preprocessing and data cleaning were performed on trajectory data,point of interest data and road network data.In particular,abnormal value detection and stay point extraction were performed on trajectory data,and POI data was reclassified.Considering street block is the basic unit of urban morphology and urban cognition,this paper divided the research area based on the road network to generate traffic analysis zones as the study scale.Secondly,natural language processing models were used to extract the latent semantic features of multi-source geographic data.Each geographic entity in geographic data was regared as a vocabulary in the text,and corpora of training word vector models were constructed.High-dimensional semantic feature vectors representing the potential information of spatial distribution and spatial interaction were obtained.And these feature vectors were weighted average,integrated and fused effectively to each traffic analysis zones.Finally,the random forest method was used to quantify the urban land use in Beijing.According to socioeconomic attributes and functions,this paper divided urban land into five types: commercial,residential,working,recreation and public service land.Based on the random forest algorithm,the relationship between the geographic semantic features of trajectory and point of interest data and the mixed urban functional areas was modeled,analyzed and evaluated.The intensity and mixing degree of various types of urban land use were quantified.The results showed that using emerging natural language processing tools can effectively vectorize the spatial structure information and dynamic spatial interaction information in the city space.These geographic semantic features not only were easy to update,but also identified more complex and diverse urban functional land.Compared with only using a single data source,this method improved the recognition accuracy of urban functional land,and can obatin higher accuracy in urban land use classification(OA = 0.7215,Kappa = 0.6758).A high average proportion accuracy can also be obtained when quantifying the proportion of various urban functional areas(PA= 66.0%).In addition,the spatial distribution of mixed urban land uses had certain reference significance for Beijing's urban planning,and can be used to help urban planners monitor changes in urban land use and evaluate the impact of urban planning schemes in real time.