Comparision Of Classification Method And Classification Accuracy Assisted With Texture

With increasing of the resolution, the remote sensing image has richer and clearer spatial information and spatial geometry structure information and texture information, so it was widely used in land classification; the traditional classification method based on spectral information of pixel has already can’t make full use of this information, at the same time,"different thing with the same spectrum " and "same thing with the different spectrum " means that classification accuracy only using spectral information of original data is limited, has cannot satisfy the increasing application demand on the modern time. As the basic information of the remote sensing image which is spectral and texture and also is the basic feature of remote sensing image classification, the fusion of texture characteristic information to the original remote sensing image can effectively extract rich information of image, plays an important role for accurate recognition of object and classification.The data sources is sectional Quick-Bird image from of "a picture" image covering HunYuan county in Shanxi Province, using co-occurrence gray level matrix(CGLM) based on statistics texture analysis method to extract the texture feature of target image, make the image of the first principal component as the texture feature extraction object, analysis spectral spectrum and texture of the typical land use class, in order to crease the differentiation of the class, the research make construction parameters based on the former analysis, then classify the image which compound the texture spectrum information and spectrum information using the maximum likelihood and decision tree method, compare the classification results of two classification method, analysis the influence of the kinds of information on classification accuracy, In this paper, The main contents and conclusions of this paper can be summarized as below:(1) According to land use classes of the research area is certainly connected with the slope gradient, using feature image and slope map to divide the study area into the valley and plain, then construct the parameter of bands Bmix to increase differentiation between grassland and agriculture land, then composite the information to the images which will be classified.(2) After a series of processing and parameter adjustment, extract the parameters of Texture feature:mean, variance and entropy, contrast, etc, on target images with the of co-occurrence gray level matrix method, and build a new combination parameters MC, composite the information to the images which will be classified.(3) The analysis of the classification accuracy show that different kinds of texture feature using will lead to the different improvement to image classification accuracy on the same classification method, mainly include:the Entropy improved classification accuracy of forestland, when the Mean has no obvious improving to the accuracy of classification results, the introduction of the Variance even makes the classification accuracy decreased; when using the decision tree method in the classification, the accuracy has more big improvement than the maximum likelihood method, at the same case of using Entropy, the accuracy of maximum likelihood method increased4.3%when the decision tree increased8.1%, the decision tree classification method which using multiple source files mining data can more effectively make use of lots of auxiliary information.