基于排列熵和CSP融合的脑电信号特征提取

EEG Signal Feature Extraction Based on PE and CSP Fusion

脑电信号(EEG)是一种在医学领域应用非常广泛的生物电信号。单一的特征提取方法不能够多方面表示脑电信号特征,从而会给不同意识任务下运动想象脑电信号的分类带来一定困难。对此,提出一种基于离散小波变换(DWT)、排列熵(PE)和共空间模式算法(CSP)的特征提取方法(DWT-PECSP)。首先,采用db4小波基对原始脑电信号进行3层小波分解,根据左右手运动想象所处的频段重构出包含μ节律(8 Hz-12 Hz)和β节律(18 Hz-26 Hz)的频段信号;然后,分别计算出该频段信号的排列熵值和CSP方差作为特征量,并将这两组特征量进行组合;最后,将组合后的特征量输入到支持向量机(SVM)中进行分类识别。实验结果表明,该算法在2003年脑机接口竞赛的标准数据集(DataSet Ⅲ)分类上获得了较高的分类准确率(91.43%),均高于单一提取排列熵特征的准确率(71.42%)和CSP方差特征的准确率(85.71%)。通过对比近年来其他文献的特征提取方法,验证了DWT-PECSP算法能够更有效地提取运动想象脑电特征。

The electroencephalogram(EEG) signal is a kind of bioelectric signal widely used in medical field. A single feature extraction method can’t represent EEG features in many ways, which will bring some difficulties to the classification of motor imagery EEG signals under different conscious tasks. For this, a feature extraction method(DWT-PECSP) based on discrete wavelet transform(DWT),permutation entropy(PE) and common space pattern(CSP) is proposed. Firstly, the original EEG signal is decomposed by three-layer wavelet based on db4 wavelet basis, and the frequency band signal containing μ rhythm(8 Hz-12 Hz) and β rhythm(18 Hz-26 Hz) is reconstructed according to the frequency band of left and right hand motion imagination. Then, the permutation entropy and CSP variance of the frequency band signals are calculated as feature quantities respectively, and these two sets of feature quantities are combined. Finally, the combined features are input into support vector machine(SVM) for classification and recognition. The experimental results show that the proposed algorithm achieves high classification accuracy(91.43%) on the standard DataSet Ⅲ of the Brain-Computer Interface Competition in 2003,which is higher than that of extracting permutation entropy feature only(71.42%) and CSP variance feature only(85.71%). By comparing the feature extraction methods of other literatures in recent years, it is verified that DWT-PECSP algorithm can extract EEG features of motor imagination more effectively.