The Classification Of Multispectral Remote Sensing Image Based On Least Squares Support Vector Machine

SVMs (short for Support vector machines)is developing from the statistical learning theory ,it is a general learning method and it is based on the structural risk minimization principle and overcomes the disadvantages of traditional classified methods which are based on the experiential risk minimization principle. But, SVMs is high computation complexity , large memory demanding and very difficult to use for large scale data sets, however, remote sensing image classification usually is large data sets, in order to improve the disadvantages, this paper proposes that one should use LS-SVM(short for the least squares support vector machines )and its improved algorithms—weighted LS-SVM and sparseness LS-SVM to classify the multispectral remote sensing image, then good classification results are gotten.This paper focuses on the principle and correlative conceptions about LS-SVM, and compares it with the traditional classified methods and SVMs through some experiments, demonstrates that it is feasible and valid. At the same time, the paper primarily probes into some peculiar problems such as multi-classification, the train examples selection and the model parameter selection in the classification of multispectral remote sensing image, then gets a conclusion that LS-SVM and its improved algorithms are good performance to classify.