基于SVM的脑功能分类与识别方法研究

Classification and recognition of brain function based on support vector machine

目的探讨SVM分类器用于脑功能识别的可行性、有效性与优越性,为脑电信号处理及功能识别提供一种新的途径和参考。方法对400组实测的正常人在睁眼和闭眼两种状态下的脑电信号,选取四种核函数分别构造四种SVM分类器对上述两种状态下的脑功能进行分类识别,从不同角度深入分析和比较讨论了由四种核函数构造的SVM分类器性能,并提出了脑电信号特征参数从低维到高维的组合变换新方法。结果由RBF核函数构造的SVM分类器最为适合脑功能的分类识别,正识率最高可达96%。结论支持向量机的方法用于脑电信号处理及功能模式识别是可行的、有效的、并初步表现出了优越的性能。

Objective To discuss the feasibility, validity and advantage of using SVM classifier in cerebal function recognition, and to provide a new method of EEG signal processing and function recognition. Methods The performance of the SVM classifiers which was constructed by four kinds of kernel function was compared and discussed in detail from different aspect using the feature parameter of real measured EEG signal. Four hundreds groups of actual normal human＇s EEG signal in state of eyes-closed and eyes-opened were adopted in our experiment. A novel method for feature parameters transform from low dimensional to high dimensional was put forward. Results The experiment result indicated that the SVM classifier with RBF kernel was suitable to the classification and recognition of EEG signal and the highest ratio of correct recognition reached 96%. Conclusion SVM is feasible, effective and with predominant performance in EEG signal processing and brain function pattern recognition.