Urban Space Extraction,Evolution And Sustainable Evaluation Based On Machine Learning

As one of the core issues in urban research,the degree of urbanization can reflect the degree of social and economic development of a city,and it is also the only way for each country to realize industrialization and modernization.The 18 th National Congress of the Communist Party of China clearly put forward the concept of a new type of urbanization,which requires the realization of a comprehensive,coordinated and sustainable development of society,economy,and ecological environment.Therefore,under this background,how to sort out,analyze and evaluate the problems in urbanization in a timely manner and promote the sustainable development of urbanization is an important topic to be studied urgently at this stage in China.With the continuous increase of high-resolution images,machine learning represented by deep learning has developed rapidly,and big data analysis technology has continuously moved to industry applications.These advanced technologies have been combined with urbanization monitoring,evolutionary analysis,and sustainable evaluation.The strategic choice to promote the sustainable development of urbanization is very meaningful and is urgently needed for the development of urbanization.Therefore,in this paper,after combing the existing research frameworks at home and abroad,the research work of urban space extraction,evolution and sustainable evaluation based on machine learning is carried out.The main research contents are as follows:(1)Using machine learning method to realize the spatial extraction of urban land information.Based on the data of geographical conditions and high-scale topographic map,this paper makes different sample making strategies and constructs the sample library of urban space land classification in accordance with the requirements of in-depth learning model training.In the traditional supervised classification framework,by combining the local features of SIFT description and the global features of spatial pyramid description,a scene classification algorithm based on the combination of spatial BOW model and Gaussian process is carried out.According to the characteristics of high-resolution satellite remote sensing image,the recognition requirements of small targets and multi-scale buildings and the practicability of algorithm,based on the existing SSD recognition framework and the network model with strong feature extraction ability,a building recognition method based on the size conversion detection network is proposed to realize the high-precision recognition of building targets.At the same time,according to the extraction requirements of complex urban space land,based on the fusion decision rules proposed by the probability of scene classification and machine learning principle,an integrated classification algorithm of urban space land based on scene classification and building recognition is proposed.(2)The research on the urban spatial evolution and its driving force of Tianjin from2008 to 2016 is carried out.It mainly includes three aspects: 1)proposed a three-dimensional analysis framework of urban spatial expansion based on "expansion scale expansion type expansion direction";2)Based on the index of urban spatial analysis of comprehensive compactness,the evolution of urban form in Tianjin is studied;3)This paper proposes and implements the analysis method of urban spatial driving force based on lifting tree to identify the driving factors of Tianjin’s spatial evolution for many years.The results show that: 1)the expansion intensity and speed are as follows: six core districts <Binhai New Area < four districts insides the city < five districts around the city;the expansion range is as follows: six core districts < four districts insides the city and five districts around the city < Binhai new area;2)The overall development direction of each region in Tianjin is basically the same,and the standard deviation ellipse is mostly narrow and long,with obvious orientation characteristics.Among them,the standard deviation ellipse in the peripheral five districts shows the transformation trend from narrow and long to standard,and there is a certain polycentric development trend in Binhai New area and the four districts insides the city;3)The comprehensive co mpactness of Jizhou district and Baodi district is always at a high level,which is related to the local ecological protection policy.The comprehensive compactness of Binhai New Area is always at a low level,and the decline of the comprehensive compactness of Wuqing district is the most obvious;4)the coupling degree of the scale and compactness of six core districts in the city is high,and the coupling degree of four districts around the city,five peripheral districts and Binhai New Area is low During the study period,there was no significant change in the coupling degree of each district in Tianjin;5)The spatial expansion of Tianjin is the result of many factors,among which GDP,population density and actual direct use of foreign capital are the main driving forces of urban expansion of Tianjin;6)the main influencing factors of urban morphology change of Tianjin are population factors.Specifically,the impact of population density,agricultural population and non-agricultural population on urban spatial compactness is far greater than other socio-economic factors.(3)The evaluation index system and BP neural network evaluation model of four-dimensional urban sustainable development based on "economy,technology,society and environment" are established.Through the analysis of the dimension of urban sustainable development,the four-dimensional evaluation theory framework of urban sustainable development is constructed.Based on this,19 indicators of sustainable development evaluation are selected by using the top-down top-level design method and literature extraction method.Then,entropy method is used to determine the index weight of each subsystem.The main reason is that it can objectively evaluate the comprehensive level of sustainable development of urban agglomerations.At present,in the academic and practical application,there are many questions about the scientificity of the evaluation index selection and its weight,and the accuracy of the final evaluation results.Be aiming at the above problems,the method of BP neural network is proposed to evaluate the sustainable development of cities.Starting from the four dimensions of sustainable development,19 indicators including four dimensions are selected by literature extraction method,then the input layer,hidden layer and output layer are determined,and a 19 × 5 ×1 BP neural network model is constructed.(4)The sustainable development evaluation and empirical analysis of Tianjin have been realized.Using the relevant statistical data of Tianjin statistical yearbook and Tianjin Science and Technology Yearbook from 2008 to 2016 and the water resources,per capita green area and urbanization data extracted in this paper,the BP neural network evaluation is realized by Tensorflow package.The results show that although the realization of Tianjin’s sustainable development is good,there is still room for improvement.By analyzing the sustainable development level of each subsystem,it is found that Tianjin has a higher level of economic and social development,a more stable economy than the social system,and the level of sustainable development of science and technology has been raised year by year to a higher level.However,the development level of Tianjin’s environmental system is general,and it has not improved significantly from 2008 to 2016.The sustainable development level of environment has obviously reduced its comprehensive strength of sustainable development.The thesis includes three innovations.(1)Based on the classification of urban construction land,the spatial extraction of urban land information based on machine learning method is proposed.As the open-source database represented by Image Net is difficult to be directly used in remote sensing image classification,this paper constructs a building sample database that meets the requirements of deep learning model training.In the traditional supervised classification framework,the scene classification algorithm is realized by combining the local features described by SIFT and the global features described by spatial pyramid.According to the characteristics of high-resolution satellite remote sensing image,the recognition requirements of small targets and multi-scale buildings and the practicability of algorithm,a method of building recognition based on dimension transformation detection network is proposed to achieve high-precision recognition of building targets.At the same time,according to the extraction requirements of complex urban space land,an integrated urban space land classification algorithm based on scene classification and building recognition is proposed.(2)Through the three-dimensional analysis framework of urban expansion,urban spatial analysis index of comprehensive compactness and urban spatial driving force analysis method based on promotion tree,the research work on urban spatial evolution and driving force of Tianjin from 2008 to 2016 was carried out.This paper puts forward the three-dimensional analysis framework of Tianjin urban spatial expansion based on "expansion scale expansion type expansion direction",integrates the urban spatial analysis index of compactness,realizes the research on the evolution of Tianjin urban morphology,proposes and realizes the analysis method of urban spatial driving force based on lifting tree,and identifies the driving factors of Tianjin’s spatial evolution for many years.(3)From the point view of four dimensions of urban sustainable development,a neural network model based on BP neural network is constructed to evaluate the sustainable development of Tianjin,and an empirical analysis is carried out.Based on the principle of evaluation index system,19 indexes including four dimensions are selected by using literature extraction method and the availability of data is considered,and then a 19 ×5 × 1 BP neural network topology training model is constructed.Based on the relevant statistical data of Tianjin statistical yearbook and Tianjin Science and Technology Yearbook from 2008 to 2016,the data of per capita water resources,per capita green area and urbanization come from this paper,and the BP neural network evaluation is realized.The result shows that although the overall situation of Tianjin’s sustainable development is good,there is still room for improvement.Through the analysis of the sustainable development level of each subsystem,it is found that the sustainable development level of the environment has significantly reduced its comprehensive strength of sustainable development.In contrast,Tianjin has a higher level of economic and social development,a more stable economy than the social system,and the level of sustainable development of science and technology has been raised year by year to a higher level.