Urban Functional Area Identification Based On Remote Sensing Images And User Visit Data

Urban functional areas are mainly divided into residential areas,industrial areas,and commercial areas.Urban functional area recognition is dedicated to extracting the complex features of each functional area.It performs function prediction on areas with unknown functions,so as to understand the distribution of functional areas,which is of great significance to decision-making.Traditional urban functional area identification methods require manual collection of information.Its efficiency is low,and the results obtained are very one-sided.The recognition result often lags behind the actual situation of the city.The era of big data provides richer and more real-time data for the research of urban functional area identification,and further promotes the technological development in this field.Currently,the data sources that can be used to identify urban functional areas include user visit data and remote sensing image data.This paper selects the data set of the 2019 “One Belt One Road” International Big Data Competition to conduct research on the identification of urban functional areas.The work is as follows:(1)Aiming at the recognition based on remote sensing image data,a model of urban functional area recognition based on Residual Network is proposed.Use data cleaning and data enhancement methods for invalid data removal and data set expansion.Through migration learning,the built Res Net152 pre-training model is used as the basis of training to extract the depth features of the remote sensing image of the functional area.The model is optimized from optimization strategies such as learning rate and loss function,and a recognition model of urban functional areas based on remote sensing images is established.Experiments prove that the recognition algorithm is effective,and the model recognition accuracy is better than that of manual recognition.(2)For identification based on user visit data,a functional area identification model based on Dual Path Network is proposed.The text data is converted into matrix data according to 7days a week and 24 hours a day.Input the matrix into the dual-path network for training to obtain the depth characteristics of the text information,and the model achieves a better recognition effect.On the other hand,through data statistical methods,the structural features of text information are extracted,and features are grouped according to the week,month and other rules.Establish a functional area recognition model based on the Extreme Gradient Boosting algorithm,and train multiple feature groups in parallel.This model achieves a high recognition effect.(3)Both remote sensing images and user visit data have certain limitations.In remote sensing images,the similarity of urban buildings is high;in user visit data,user actions are disorderly and excessively rely on sensors to collect information.Therefore,the functional area information obtained from single-type data is relatively one-sided.In order to make up for the limitations of image modal and time series modal data,this paper proposes a multimodal fusion classifier(MFC)to identify urban functional areas.Extract the deep features of multiple single models through transfer learning.Input it and structured features into MFC for city functional area identification.On the other hand,the limit gradient boosting algorithm is used for fusion training of three single models,which is used as a comparative experiment.The experimental results verify the effectiveness of the MFC classifier.And the recognition accuracy of the MFC classifier based on the fusion of the three types of features is significantly improved compared with the single model,and its recognition effect of urban functional areas is better than other algorithms.