Research In The Definition Of Urban Fringe Area Based On Deep Neural Network

In the context of the rapid development of urbanization,the disorderly expansion of cities makes the characteristics of urban are more and more similar to the rural areas and the boundary between urban and rural areas is more and more blurred,thus leading to the emergence of urban fringe areas with both urban and rural characteristics.As the gathering place of the conflict of urban and rural spatial characteristics and the integration of matter and culture,urban fringe area has become the zone with the largest amount of spatial morphology change,the most complex integration of material and culture,and the most sensitive to external stress.Therefore,the urban fringe area has some typical characteristics,such as complexity,sensitivity,transitivity,dynamism and so on.Due to the particularity of land use in urban fringe area,it is easy to cause a series of regional contradictions and conflicts caused by urban's sprawling.Especially,the uncertainty of the spatial scope of urban fringe areas,which leads to unclear responsibilities in management,lagging administrative management and disordered land development and utilization.Therefore,having an objective and scientific method to define the urban fringe area accurately,which plays a basic role in the planning of urban and rural social and economic development,solving a series of urban diseases caused by urbanization,and optimizing the management of urban and rural.This paper focuses on a basic issue in the research of urban fringe area——identification of urban fringe.The typical city as the main research area in China,Guangzhou,Shenzhen,Hangzhou,Wuhan,Chengdu,Xi'an have been selected.In view of nature and social economy,with the foundation of big multi-source data and machine learning algorithm,to build a model to identify urban fringe accurately,universality and objectivity.Comparing the accuracy and stability of different algorithm models about DNN and SVM,and discussing the influence of different data sources to the results.Finally,using the multi-source and DNN model to identify the urban fringe of six cities in China.In the results of six research cities,regional spatial structure development differences and urban social and economic development status of 6 typical cities in China are discussed.The results show that:(1)Building DNN(Deep Neural Networks)model to identify urban fringe areas.In the view of nature and social economy,this paper has selected land use,population and big POI data as the main data source.Meanwhile,after reading much literatures,the data of landscape disorder,population,catering industry,companies,hotels,financial services,scientific research and education are selected as important indicators.Finally,combining with DNN model and multi-source data to build a model to identify urban fringe area in the basic of enough sample about city,urban fringe and rural area.(2)Comparing with the model of DNN and SVM in the identification of urban fringe.Comparing with the model of DNN and SVM in the identification of urban fringe based on the same index data.The difference in the results of urban fringe area between DNN and SVM model was verifying the superiority of DNN model in the identification of urban fringe area.(3)Comparing with the influence of different data sources on the identification of urban fringe.On the basis of the DNN model for identifying urban fringe,the results of identifying urban fringe from remote sensing data,POI data and multi-source data source were compared and analyzed,which is verifying the advantages of multi-source fusion data in the identification of urban fringe.(4)Comparing the results of urban fringe areas in six typical Chinese cities.On the basis of multi-source big data and DNN model,this paper analyzes and compares the results of urban fringe of six typical Chinese cities which are Guangzhou,Shenzhen,Hangzhou,Wuhan,Chengdu,Xi 'an,in order to clarify the differences in urban structure and social economic development of different cities.