支持向量机在脑电信号分类中的应用

Application of SVM in EEG signal classification

首先采用小波变换提取精神分裂症与健康人的脑电信号频率和空间的能量特征,然后用基于统计学习理论的支持向量机(SVM)分类器进行训练和分类测试,并比较了不同核函数和参数对脑电信号分类正确率的影响,最后与RBF神经网络的分类能力进行了实验比较。试验结果表明,利用基于支持向量机和能量特征的方法实现对脑电信号的分类可以取得理想的效果,精神分裂症患者和健康人的16导脑电信号在能量特征上表现出较高的模式可分性。这种分类方法在精神分裂症患者的病理诊断中具有一定的应用价值。

At first some energy features related to the frequency bands and special distribution are extracted, Then a classifier based on Support Vector Machines （SVM） was designed. To optimize the classifier, different kernel functions and parameters were discussed. The performance of classifier was compared with a RBF Neural Network classifier. The result indicates that the ideal accuracy can be achieved by the SVM and wavelet energy method in EEG classification. The schizophrenia can be separated from healthy through 16-channel＇s EEG. The research takes important practical value in the schizophrenic diagnose.