Detection Of Land Use Change In Urban Villages Based On High Resolution Images

With the development of China's economy,China's urban development is rapid,and the city or the city within the city edge of the village,relatively speaking,the pace of development will be relatively slow.There are a series of problems such as poor living environment and incomplete land use information.In recent years,Xi'an has undergone tremendous changes,in order to promote urban development,High-tech zones,Chanba District,Chang'an District,part of the village demolition and reconstruction.This article takes the city of Chang'an District of Xi'an as an example,for these issues,to investigate.In this paper,we use the high grade one images made in China in 2013 and 2015,combined with the advantages of high resolution images,the land use changes of villages in the study area were detected,and the change information was extracted automatically and analyzed by time and space.This paper adopts the object-oriented classification method of traditional change detection,first,a typical support vector machine is used to detect the change after classification average overall classification accuracy of two images was 70.40%,the change detection accuracy is 49.55%,the classification accuracy is low,resulting in the change detection accuracy is low,and serious error;then use the object-oriented change detection and classification method,the classification method based on support vector machine method,the average overall classification accuracy was 78.69% and the accuracy of change detection is 63.46%,but the error is still obvious.Since the object-oriented method has a single scale problem,scale selection can not use all the ground features,thus seeking a new solution.In order to alleviate the above problems,this paper introduces another method of post classification change detection,which uses multi scale learning method and relates hierarchical conditional random field method.First,the high one image into 500*500 blocks,verify 8 training samples per issue,use LBP,Texton,Spectral feature,the average value obtained in 2013 and 2015,the overall classification accuracy is 91.03%,can be seen using this method is feasible,then classify the whole image,the average overall classification accuracy is 88.90%,compared to the object-oriented classification method to improve the accuracy of change detection was 9.21%,79.01%,using the method of visible change detection is more effective.Then,the change detection in the study area is analyzed.It can be concluded from the data of 2013 and 2015 as part of the demolition of villages,but there are also part of the reconstruction,the construction land has decreased,but the decrease is not much;more water to reduce;little change in vegetation and open space,but also changed.