基于总体经验模态分解的多类特征的运动想象脑电识别方法研究

人的脑电信号（Electroencephalogram，EEG）复杂且具有非线性及非平稳性的特点使其不易分析处理，其识别效果也依赖于数据集的不同，而表现不稳定.本文中应用的总体经验模态分解（Ensemble empirical mode decomposition，EEMD）是一种具有强自适应性的信号处理方法，其在时频域展现的良好分辨率特别适合脑电识别任务处理.本文提出利用EEMD分解后得到的较具影响能力的固有模态函数（Intrinsic mode functions，IMFs），利用希尔伯特变换提取边际谱（Marginal spectrum，MS）及瞬时能谱（Instantaneous energy spectrum，IES）时频特征，同时通过加窗的方法提取非线性动力学特征近似熵特征，利用线性判别分类器（Linear discriminant analysis，LDA）作为分类器，实验结果得出，对于被试S2和被试S3可达到识别率分别为79.60%和87.77%，实验中9名被试的平均识别率为82.74%，得到平均识别率也高于近期使用相同数据集文献的其他方法.

Multiple Feature Extraction Based on Ensemble Empirical Mode Decomposition for Motor Imagery EEG Recognition Tasks

EEG signals are complicated as well as nonlinear and non-stationary, which make them hard to analyze. Recognition result is dependent on the datasets selected, and is not stable. The ensemble empirical mode decomposition (EEMD) as a kind of adaptive signal processing method is used for motor imagery recognition tasks because of its good decomposition resolution. An efficient EEMD-based feature extraction scheme is presented, which combines the Hilbert marginal spectrum (MS) and instantaneous energy spectrum (IES) features with window-added EEMD-based approximate entropy (ApEn) features. The impactful factors of IMFs and frequency bands are selected for the features as well. A linear discriminant analysis (LDA) classifier is designed for classifyication. The method is tested on nine subjects. The result shows that the proposed feature combination is competitive in recognition rate with other methods on the same dataset. The maximal classification accuracy for S2 and S3 can reach 79.60% and 87.77%, respectively. The mean accuracy of nine subjects is 82.74%. The average recognition rate obtained is superior to other methods on the same datasets.