Study On Method Of Urban Area Land Cover Classification Based On ZY-3

As the rapid development of economy and urbanization, the great change has also taken place in land cover. Such as agricultural land reduction, ecological degradation, environmental pollution, expansion of urbanization, etc. These vatiations not only changed the character of environment, but also had an effect on the character of environment, also has impact on the social economy, ecosystem and hydrological system. The classification could be as data foundation of rational use of land, also could provide scientific basis for planning land use and constituting sustainable development policy. So, the classification method of land cover has always been the core and focous in study on global variation and geological field. Land cover classification with remote sensing technology to achieve the extraction of land cover information had become a significant topic among remot sensing application. With the development of remote sensing technology, the higher resolution of remote sensing image, the richer feature information that image negatived. But, the traditional method cuoldn’t make full use of spectral, space, texture information of high resolution image, because it was based on pixel. The results of traditional classification method had an influence of spectral, what leaded to exist a lot of “noise”. So, the accuracy of classification result can not meet the demand of industry. Therefore, the objected-oriented classification method arises at the moment for that high resolution remote sensing image data effectively applied to the surface cover classification research.The process of Object-oriented classification was as follows: Firstly, the computer divided image into objects based on segementation scale. Secondly, detection and extraction a variety of characteristics what can describe target object would be done according to the features of the spectral, texture and space characteristics. Finally, the method of fuzzy classification was applied to the classification of remote sensing image and extraction of information. This method had a very strong sntinoise ability because that it based on Object. It also could make up for the shortage of the method based on pixel level effectively. But, the current classification method based on object covered two difficult challenges: the determination of the optimal segmentation scale and the optimization of the festure space. With the support of ERDAS and MATLAB platform, the article did some reseach on these two method optimization from the core problems. The main research contents and results were as follows:(1) Explore optimal proportion of band recombination of ZY-3.This article did an experiment that recombine band G with band G and band Nir in a different proportion. It proved that the optimal proportion is 9:1, and the output imagehad an appropriate illumination and contrast, clear texture,and it contained a largest amount of information. So, this result was well for remote sensing interpretation.(2) Analyze the best fusion algorithm of ZY-3: Firstly, Based on ZY-3 remote sensing image, the experiment was carried out by method of PCA, Mutiplity, Gram-Schmidt, Ehlers, Wavelet transform. Eight indicators that grayscale average, relative standard deviation, the root mean square error, spatial frequency, spatial frequencies,the information entropy, cross entropy and correlation coefficient were used to judge and analyzed the fusion results objevtively. The experiment results showed that PCA fusion method greatly improved the image quality. The result of PCA fusion algorithm had a suitable brightness, clearly texture. It contained a large amount of information and an well image level. It could provide a data foundation for land cover classification.(3) Do a survey on the question to determimatin of the optimal segmentation scale. The article did a deep research. It compared the different method,analyzed the different superior function indicators what contained the max-area, local variance, RMAS, homogeneity and heterogeneity. Then, the idea of homogeneity and heterogeneity was considered, an evaluating index include intracontrast index, intra-entropy and standard deviation of spectrum which is called ENACS index was putted forward on the basis that it taken account the between objects after the characteristics of spectral information and the image segmentation object within the spectral and spatial characteristics comprehensively. The result showed that after the improved index the reliable optimal segmentation scale could be obtained conveniently, it was broader in scope.(4) A improved algorithm for optimal feature selection was proposed. Based on the algorithm of object-oriented method, which is named as Separation Threshold Method(SEath), the paper adopts the search algorithm based on J-M distance criteria to choose the feature. The improved algorithm considered the de-correlated feature and the distance in internal class to get a new indicator for the optimal characteristics selection. And this paper searched the optimal feature subset by Plus-L Minus-R Selection algorirhm(LRS). The improved algorithm not only could improve the search efficiency, also could avoid the local optimal feature subset.(5) Study the object-oriented classification technology of high resolution remote sensing image. Based on ZY-3 satellite remote sensing image, the article did an experiment about land cover classification according to the decision rule tree that built by two algorithm what the optimal segmentation scale decision and optimal feature subset search. The experimental results showed that the object-oriented classification method could effectively solve the "same thing different spectrum" "foreign body with spectrum" phenomenon; Second, BAI building index could effectively extract the vegetation, the near infrared band mean could extract waters well; At the same time, using the improved optimization method for the optimal feature selection, the Kappa coefficient of overall improved from 0.75 to 0.88, the overall accuracy of information extraction results from the original 83.00% to 91.02%.