Support Vector Machine Applications In Signal Analysis And Identification Of Brain Function

The Support Vector Machine algorithm which taking the kernel function as foundation is based on the VC Dimension theory of statistical learning and the principle of Structural Risk Minimization. It looks forward a compromise between the model complexity (the learning accuracy to the specified training sample) and learning capability (the faultless recognition to the random sample) according to the finite sample information. Since the SVM was put forward by Vapnik and his colleagues in the middle of 1990s, the algorithm and application of SVM have been acquiring a rapid development.In this paper, we analyze thoroughly the principle of SVM and kernel function at first. Then construct four kinds of SVM classifiers with four different kernel functions to analysis and recognize the brain function, which under different kinds of behavior. After compare the capability of the different classifiers we could find out a kernel function which adapt to extract and classify brain function. Finally in order to discuss and prove the feasibility and validity of the SVM method in the analysis and recognition of brain functional signal, we use the original EEG signals which have two kinds of behavior to perform four experiment, the results and conclusion are as follows:(1) The sample values of the original EEG signals in time domain of single channel are taken as the feature value, and the result shows that the SVM classifier based on RBF kernel function makes obvious higher correct rate of classification than others.(2) The feature parameters of several single-channels are combined as feature parameters of multi-channels for the recognition, the result shows that feature parameters of multi-channels contain more information of EEG signals and gains better recognition and higher reliability.(3) The grey models are constructed to the extract of features which are to be classified, the results also indicate that the SVM classifier based on RBF kernel function makes obvious higher correct rate of classification than others.(4) If we increase the dimension of the features through combine the extracted feature parameters of grey models, the rate of correct classification will be also increases.