Abstract: The Electroencephalogram（EEG） of mouse model of epilepsy in normal and epileptic status is collected to study the automatic classification of epileptic EEG. The noise- and artifact-attenuated EEG is wavelet-transformed, and the linear feature（standard deviation） and the nonlinear feature（sample entropy） are then extracted for the EEG signals and those wavelet coefficients related to the characteristic waveforms of epileptic EEG. Classification is implemented using support vector machine with above individual features and their combinations. The classification accuracy based on the standard deviation and sample entropy of EEG signals are 59.1% and 58.00%, respectively.The accuracy increases to 86.60% or 88.60%, when the standard deviations or sample entropies of relevant wavelet coefficients are used as input features. After combining both types of features, the classification accuracy is 99.80%. These results show that wavelet coefficient features fusion significantly improves the classification accuracy, achieving effective classification of mouse epileptic EEG.

Key words: epileptic mice model, wavelet transform, feature fusion, support vector machine

摘要： 采集癫痫小鼠模型在常态与致癫状态下的脑电信号以研究其癫痫脑电的自动分类。对经过噪声和伪迹消除预处理的脑电信号进行小波变换，获得不同频率子带的小波系数，对脑电信号及与癫痫特征波相关的小波系数提取相应的线性特征（标准差）和非线性特征（样本熵）；基于这些特征及其组合使用支持向量机分类器实现分类。实验发现基于小鼠脑电本身的标准差和样本熵的分类正确率分别为59.10%和58.00%；而融合各相关小波系数的标准差或样本熵，分类正确率分别达到86.60%和88.60%；融合全部相关小波系数的线性和非线性特征后分类正确率为99.80%。这些结果说明基于小波系数特征融合的分类算法性能有显著提升，能有效实现小鼠癫痫脑电的自动分类。

关键词: 癫痫小鼠模型, 小波变换, 特征融合, 支持向量机