| In recent years,great changes have taken place in the remote sensing industry and with the rapid development of sensor technology and electromagnetic wave receiving and processing technology,a series of high spatial resolution satellite sensors emerged during this period.how to make full use of high resolution remote sensing images to provide real-time and accurate land cover information for China’s land use,urban and rural planning,land change monitoring,military and other industries is very important.Remote sensing image classification is the most labor-saving,accurate and important way to obtain land cover information,and its technical progress has been paid much attention to.support vector machine(SVM)algorithm has achieved considerable results in remote sensing image classification..In this paper,based on the analysis of the status quo of support vector machine(SVM)classification of remote sensing images,making reference to the second national land use survey classification system to formulate classification standards of research areas,we study the SVM classification of high-resolution remote sensing images based on multi-feature.This paper provides some reference value for the classification of high-resolution remote sensing images based on support vector machine to some extent,and also provides some technical reference methods for the extraction of land cover information.The main research contents and results are as follows:(1)We using GF-2 remote sensing image near Yanqi lake in Huairou district of Beijing as the research object,based on the basic theory of SVM and the basic characteristics of high spatial resolution of experimental data in the study area,the classification of high-resolution remote sensing image was studied by fusing the spectral values of image band and NDVI,and the four texture features of variance,entropy,dissimilarity and second moment,And compared with maximum likelihood classification and neural network classification.(2)As far as the classification results of GF-2 images in the research area are concerned,through the comparison and analysis of various precision indexes and classification diagrams,the classification method based on multi-feature has obtained the best classification effect.the overall precision is 95.67 %,kappa coefficient is 0.9267.(3)Compared with other classification methods,the SVM classification accuracy is the best,and the degree of patch fragmentation is smaller than other methods,as far as the GF2 image classification results in the research area are concerned,whether the image band spectral values are used alone or combined with image spectral and texture features.In addition,the maximum likelihood classification and SVM classification in the comprehensive consideration of image spectral information NDVI and texture information(variance,entropy,dissimilarity,second moment and PCA to select the main component)when the classification accuracy is better than the single use of four bands of remote sensing image classification accuracy;However,the accuracy of neural network decreases after incorporating multiple features. |
