Research On Classification Of Essential Urban Land Use Categories(EULUC) In The Central City Of Chongqing

Cities are closely related to human life.With the continuous development of my country's urbanization process,understanding the distribution of basic urban land use types has an important role in the overall urban renewal tasks such as the improvement of urban spatial structure,the protection of urban history and culture,the shaping of urban features,and the construction of residential communities and urban infrastructure.important meaning.This paper takes the urban built-up area of the central city of Chongqing as the research area,and studies the basic urban land use classification method based on multi-source data.Aiming at the current problems and deficiencies in this field,from determining the optimal urban land segmentation method,constructs suitable The characteristics of urban land use classification in mountainous cities and three aspects of determining the optimal basic urban land use type classification method are discussed.Get the following conclusions:First,by comparing the four segmentation methods of road network segmentation,multi-scale segmentation,LSMS segmentation and comprehensive segmentation,the optimal segmentation method of urban land parcels is determined.The comprehensive segmentation method that combines road network segmentation and LSMS segmentation effectively solves the problems of insufficient suburban land segmentation in road network segmentation and irregular urban land segmentation caused by LSMS segmentation.The segmentation results of urban land parcels obtained by comprehensive segmentation are closer to the morphological characteristics of real blocks,which is an ideal method for segmentation of urban land parcels.Secondly,the SEATH algorithm is used to optimize the classification features extracted from multi-source data,and the classification features suitable for the classification of basic urban land use types in mountainous cities are obtained.In the primary and secondary classification systems,the classification features extracted from POI,buildings,and visible light remote sensing data have a high degree of separation.Synthetic aperture radar,night light remote sensing,population number,and Tencent location data have improved with the refinement of the classification system.However,there is still a gap compared with other data.After analyzing the contribution of the optimal classification features,it is found that visible light remote sensing,POI,and buildings are the main data in the classification of basic urban land use types in mountain cities.And with the refinement of the classification system,the role of POI data has been enhanced.Finally,by comparing K-mean clustering analysis,supervised classification based on random forest,supervised classification based on support vector machine and deep learning classification based on Tensorflow framework,K-mean clustering analysis requires human interpretation to determine Class type,leading to more confusion in the classification results.Deep learning classification through experiments found that it does not fully utilize the role of POI,building attributes and other classification features.It is not effective when classifying urban land and land types with more complex and diverse spectral information.It requires more spatial resolution of remote sensing images.High,leading to unsatisfactory classification results.Compared with support vector machines,random forests have higher classification results.The overall classification accuracy of the primary classification system of random forest is 85.47%,the Kappa coefficient is 0.8,the overall classification accuracy of the secondary classification system is 79.34%,and the Kappa coefficient is 0.77.By comparing the classification results,it is determined that random forest is the best classification method.