A High Spatial Resolution Remote Sensing Image Classification Study Based On SVM Algorithm And Application Thereof

The rapid development of the remote sensing technology provides convenience for us to obtain more abundant information of the earth’s surface. Especially in recent years, with the appearance of high spatial resolution remote sensing satellites, it further provides us the geometric information and the texture information except the spectral information. In such circumstances, if we only use the spectral information to process the classification problem, it will result in not only wasting the redundant spatial information but also reducing the classification accuracy. Therefore, the classification problem of high spatial resolution remote sensing image should be processed by making full use of the space information.Support Vector Machine(SVM) algorithm is a kind of pattern classification algorithm based on the theory of Statistical Learning Theory(SLT) and the theory of Structure Risk Minimizing Principle. It has features of Computational Efficiency, Robustness and Statistical Stability. It is obviously showed that SVM has an advantage on processing the nonlinear classification problem because of their high generalization performance without the non- convergence problem of the algorithm, even under small sampling and high-dimensional condition Thus, at present, the current application field of SVM is very wide but the application of high spatial resolution remote sensing image is not. Therefore, the study of the application of SVM Algorithm in high spatial resolution remote sensing image classification has important theoretical and practical significance.This paper takes the terrain classification of the Yanxi lake area in Huairou district as a background. The main research purpose of this paper is to discuss the feasibility of classification and the critical technology of high spatial resolution remote sensing image. The more details are as follows:Firstly, the paper summarizes the fundamentals of high resolution remote sensing image classification and SVM Algorithm based on previous study, which are the theoretical basis of high resolution remote sensing image classification by using SVM Algorithm This paper presents the SVM algorithm framework of C-SVC type for solving the classification problem of 4 band high spatial resolution remote sensing image. In order to improve classification result, this paper tries to combine texture information and spectral information to train the Support Vector Machine to classify. In evaluation of the classification results, the paper respectively discusses the influence of the parameters of SVM and the three kinds of kernel functions(Radial Basis kernel, Polynomial kernel and Linear kernel) on classification results and time performance of classification. This thesis compares K-means unsupervised classification, Minimum Distance and Maximum Likelihood Estimation supervised classification with SVM Algorithm and proves that SVM algorithm has a better representation and predominance at accuracy rate and generalization performance. It is feasible to use the SVM algorithm in terrain classification of high spatial resolution remote sensing image practically and efficiently. In the end, the parallel optimization of remote sensing image data is implemented to improve time performance of classification.