Research Remote Image Classification Based On Support Vector Machine

Image classification is an important means to get information by human, The traditional classification methods are based on the inductive principle of empirical risk minimization. It is achieved optimal performance, when the number of the training samples approach infinity. However, the training samples of the remote sensing are limited. The traditional classification method can't achieve the desire accuracy, when the training samples are low.Compared with the traditional statistics, the statistical learning theory is a machine learning rules, which is a special case of low sample. Support Vector Machine (SVM) is based on the statistical learning theory, overcame the shortcomings of the neural network classifier and the traditional statistical classification methods, has a high generalization ability.SVM is a new technology used in data mining, a new tool solving the problem of computer learning by means of optimization methods. The unclear mapping method is applied which is overcame the difficulties of the dimension curse and the over learning. The nonlinear problem is solved better. Compared with the traditional neural network method, it reflects the minimization principle of the structural risk. It is not only a simple structure and a strong generalization, but also solved the practical problem that is solving the low sample high-dimensional data and the local minima. It is widely used by its easy used, stability and a relatively high accuracy.Selecting the appropriate map, mapping the nonlinear sample to a high dimensional space, creating an optimal separating hyper plane with a low VC dimension in a high dimensional space is the core idea of SVM. By considering the empirical risk and the size of confidence interval, it is acquired the classification function in terms of the principle of the minimization structural risk to find the best compromise.In this paper, pilot site is the researched area. The SVM algorithm framework is built, which is used by Polynomial Kernel, Radial Basis Kernel, Sigmoid Kernel function and Linear Kernel. The classification experiment is used by the four kernel function. In this paper, it is analyzed the SVM algorithm theory and achieved the algorithm by the C++.In order to check the SVM classification result, Minimum Distance and Maximum Likelihood Estimation methods are compared with it. There are a lot of advantages in accuracy and generalization ability, so it is feasible in Image classification.