Research On Urban Functional Area Recognition Based On Multi-Source And Multi-Feature Information

With the rapid development of urbanization,the urban development model of "incremental expansion" is difficult to maintain,and the new trend of urban development will focus on the refined spatial "stock tapping".Therefore,the detailed and correct identification of urban functional areas is of great significance to enhance urban planning,improve urban order and promote sustainable urban development.Traditional methods of classifying urban functional areas are limited by data sources and data scale,and often have problems such as strong subjectivity,high labor cost and slow update speed.With the development of remote sensing technology and the improvement of accessibility of multi-source big data,scholars have explored the use of remote sensing images,Point of Interest(POI)and other data for automatic urban functional area identification,and have made great progress.However,these studies often ignore the validity of building data in feature mining,and the research results generally lack the focus on functional mixed areas.However,buildings,as the main places of human activities,can reflect the three-dimensional urban landscape and regional functions to a certain extent,and the joint application with building data can make urban zoning more significant;moreover,since urban areas are complex and heterogeneous,and the functional categories of internal parcels are not completely pure,considering mixed functional attributes will be more in line with the real situation.Therefore,this study takes two provincial capitals,Guangzhou and Harbin,as the study area,and proposes a method for detailed identification of urban functional areas based on image-building-POI data.Firstly,the image spectral features and texture features are extracted by principal component analysis and grayscale co-occurrence matrix,the2D/3D features of urban buildings are obtained by statistical measurements,and the POI features are obtained by kernel density estimation method.Secondly,the random forest(RF)model is used to achieve multi-feature fusion,and the functional type of the block is determined based on the RF voting probability,so as to achieve detailed and accurate identification of urban functional areas,and the accuracy is verified by error matrix and compliance scoring.To verify the effectiveness of this method of multi-feature fusion,this paper designs comparative experiments to explore the differences in recognition performance of different feature combinations,and quantitatively analyzes the specific contributions of different features to the recognition of various functional areas through RF feature importance calculation.The main contents and conclusions of the paper are as follows.(1)The functional area recognition method fusing multi-source and multi-feature information achieves high recognition accuracy in the application of two study areas.Among them,the overall accuracy(OA)of single functional areas in Guangzhou and Harbin are 83.33% and 84.37%,respectively,and the accuracy(Accuracy score,AS)of mixed functional areas are 81.67% and 80.00%,respectively.The results of functional area analysis show that the proportion and distribution of various functional areas in Guangzhou are in a more balanced development layout,while the urban functional areas in Harbin have a relatively obvious inner-outer ring circle structure,and the functional types are more focused on secondary industries.(2)Through the parallel experiments,it is found that the recognition effect of functional areas using single features is poor,and the combination of features is beneficial to the improvement of recognition accuracy,and the recognition method of fusing multifeature information in this paper can achieve better results.In the single feature experiment,the highest accuracy is obtained using building features;in the combined feature experiment,the image-building features obtain the best recognition effect.Compared with the single-feature experiments,the OA of Guangzhou and Harbin for the method in this paper improved by 20.00%～28.33% and 23.99%～32.60%,respectively,and the AS improved by 15.00%～25.00% and 25.00%～26.67%,respectively.Compared with the combined feature experiment,OA and AS in Guangzhou improved by6.67%～11.67% and 3.34%～11.67%,respectively,and OA and AS in Harbin improved by7.81%～15.63% and 5.00%～15.00%,respectively.(3)By calculating the feature importance,we found that the contribution of image features and building features in this study was relatively high,with 30.80% and 35.71%for image features in Guangzhou and Harbin,respectively;and 48.88% and 34.63% for building features in Guangzhou and Harbin,respectively.However,the mean values of factor contributions with the effect of feature dimension removed indicate that POI features have high contributions to the identification of industrial functional areas,commercial functional areas and public service areas.Overall,this study verifies the effectiveness and complementarity of image data,building data and POI data in urban functional area identification,and provides a new perspective for urban functional area identification by fusing multi-source and multifeature information.