Research And Application Of Land Use Remote Sensing Information Extraction Based On Fractal Texture

With the increasing requirements for the accuracy of image automatic classification,classification using only image spectral information is faced with the problem of limited ability of classification to distinguish features.As an important manifestation of the spatial structure information of remote sensing images,texture features are gradually paid attention to by using texture features to assist spectral information for feature classification.Fractal theory is one of the three universities in the field of nonlinear science,and the fractal dimension is consistent with the human eye’s perception of the roughness of image texture.The fractal method is used to extract image texture features and used for land use classification.However,the predecessors lacked quantitative research on the selection of fractal bands and fractal windows.This article takes the Dexing mining area in Jiangxi as a research area,uses highresolution GF1 image and medium-resolution Landsat image,based on fractal theory,uses double blanket coverage model and difference box dimension method to extract two data sources.Texture image under the window.After the texture image is combined with the original multispectral data,the method of support vector machine is used for classification.Using human-computer interactive interpretation results as a reference,determine the best fractal band and fractal window for the two types of image data.Using the best band and window combination of multi-phase Landsat data to classify the land use situation in Dexing mining area in 1998,2003,2008,2013,and 2018.This article mainly achieved the following results:(1)Using MATLAB to realize the double blanket covering model and the difference box dimension method to extract the texture features of different remote sensing images in different windows of different wave bands,and used to assist the classification of images.(2)Through experiments,the best fractal band and the best fractal window for extracting the texture features of the GF1 data and Landsat data in the study area using the double blanket coverage model and the difference box dimension method were determined.It is proved that the first principal component band is directly used as the fractal band,and the method of selecting the fractal window through the visual effect is not desirable,and the highest classification accuracy cannot be obtained.(3)Through comparison,it is found that for GF1 data,the texture image extracted by the double blanket coverage model is more advantageous than the difference box dimension method in improving the image classification accuracy.For Landsat data,the texture features extracted by the difference box dimension method help to improve the classification accuracy.For different data sources,after adding texture features,the classification accuracy of GF1 images with fewer bands is more obvious than that of Landsat data with more bands.(4)Using land use quantity changes,structural changes and land use dynamics and other indicators,the land use changes in the study area during the past 20 years were analyzed.