Study On Land Remote Sensing Classification Of Nanchang County Based On Object-Oriented C5.0 Decision Tree

This thesis focuses on the current situation of land use at the county level under the background of social production with the steady economic development.Based on the Landsat-8 image of Nanchang County,Nanchang City,Jiangxi Province in 2018,this paper uses the object-oriented method,takes the object unit of multi-scale image segmentation as the unit,and uses C5.0 decision tree to complete the first and second level land classification of Nanchang County,and extracts land classification information.The main methods and experiments adopted in this paper include:Firstly,using multi-scale segmentation method,remote sensing images are segmented into entity units corresponding to actual object categories.Observe and analyze the effects of scale parameters,color factors and shape factors under different numerical values on the segmentation process,and the segmentation results under better segmentation parameters are selected as the data basis for subsequent processing.Secondly,based on the C5.0 decision tree algorithm,land classification is carried out at the first and second levels:according to the geographical environment and development of the study area,and with reference to the land classification system standard(GB/T 2100-2007),land is classified into five categories:garden land,cultivated land,construction land,water body and unused land at the first level;This paper further explores the secondary detailed classification of"construction land" under the favorable situation of stable opening policy and vigorous promotion of economic construction,and divides "construction land" into four categories:rural residential land,urban residential land,industrial land and development land.The experimental results show that compared with the pixel-oriented maximum likelihood method,the object-oriented C5.0 decision tree method takes into account the traditional spectral information of pixels and uses texture,shape and other spatial features of the pixels and their adjacent pixels to classify objects.The overall accuracy of the first-level classification is 0.9104,9.450%higher than the maximum likelihood method,and the Kappa coefficient of the first-level classification is 90.03%,9.45%higher than the maximum likelihood method.The overall accuracy of the second-level classification is 0.8283,11.59%higher than that of the maximum likelihood method,the Kappa coefficient of the second-level classification is 82.80%,and 11.60%higher than that of the maximum likelihood method.However,with the same object-oriented thinking and CART decision tree model,the classification effect of C5.0 decision tree algorithm is still slightly higher than that of CART decision tree.The overall accuracy of the first-level classification is 0.9104,2.37%higher than that of CART decision tree method,the Kappa coefficient of the first-level classification is 90.03%,2.37%higher than that of CART decision tree method,and the overall accuracy of the second-level classification is 0.8283,2.49%higher than that of CART decision tree method.The Kappa coefficient of the hierarchical classification is 82.80%,which is 2.48%higher than that of the CART decision tree method.The object-oriented C5.0 decision tree algorithm,as the main algorithm in the study of remote sensing image classification,can successfully extract land classification information,and its classification accuracy is higher than that of maximum likelihood method and CART decision tree,which illustrates the advantage of using C5.0 decision tree algorithm to classify County land at the first and second levels.And the first and second-level land classification rules generated by C5.0 decision tree algorithm can be expressed in the form of "rule set",which provides clear classification rules for the application research of corresponding land categories.