A Comparative Study On The Extraction Method Of Surface Features Based On Remote Sensing Index

Since the 1990 s,the land use and land cover change have become the hot topics in the whole society.Harbin is the capital of Heilongjiang Province.The economic develops rapidly,the land use and land cover change become faster in Harbinin recent years.This study is based on the center regions in Harbin city: Nangang,Daoli,Daowai,Xiangfang,Pingfang,Hulan and Songbeiby using the most widely used remote sensing classification methods as the study method to extract regional land use information.Use Landsat-8 multispectral images of the regions as the data source.Based on common supervised classification methods such as maximum likelihood method,decision tree,support vector machine(SVM)and artificial neural network,the author extracts surface features by using remote sensing characteristics and spatial index as the variable objects.Compare and analyze the precision evaluations of remote sensing classification based on pure spectra land remote sensing classification based on the index in extracting surface features to better coordinate land use of Harbin city and provide research basis for urban expansion in the future.The main results are as follows.1.Based on the pure spectral remote sensing classification methods such as maximum likelihood method,decision tree classification,support vector machine,artificial neural network to extract information of land use in Harbin.The results show that the overall classification precision evaluations are arranged in orders: the decision tree classification(83.09%)> artificial neural network(82.18%)> support vector(81.84%)> maximum likelihood(76.13%).2.Based on the pure spectral remote sensing classification methods such as maximum likelihood method,decision tree classification,support vector machine,artificial neural network,changing the characteristics of input variables to extract information of land use in Harbin.The result shows: compared with the remote sensing classification method based on index,some of the classification precision evaluations of the remote sensing classification based on pure spectral are lower,while some are higher.3.Because the index of remote sensing has different effects on the max imum likelihood method,decision tree classification,support vector machine and artificial neural network in classification precision evaluations,so selecting four groups of variables combination with the four classification methods,and then extracting land type information from the regions.The results of comparison between remote sensing classification based on pure spectral and the precision of remote sensing classification method based on the index show that the overall precision of maximum likelihood classification and Kappa coefficient increased by 6.31%,0.08 at least;7.42%,0.09 at most.The overall precision of decision tree classification minimum and Kappa coefficient increased by 0.8%,0.08 at least;2.92%,0.01 at most.The overall precision of support vector machine(SVM)and Kappa coefficient increased by 1.71%,0.03 at least;1.96%,0.03 at most.The overall precision of artificial neural network and Kappa coefficient increased by-2.56%,-0.05 at least;2.05%,0.08 at most.