基于AAR模型和累积频带能量的特征提取方法

提出了一种自适应自回归（AAR）模型参数和累积频带能量相结合的特征提取方法，该特征应用于基于运动想象脑．机接口（BCI）之中，实现左右手运动想象分类，改善BCI系统的性能．首先，对头皮EEG数据进行小波分解和重构，去除EEG中的噪声，得到不同频带的EEG数据．然后，提取EEG数据的AAR模型参数特征和不同频带的频带能量特征，提出了累积频带能量特征和AAR与累积频带能量相结合的特征提取方法，分别以AAR模型参数、频带能量、累积频带能量和AAR＋累积频带能量为特征，利用线性判别分析（LDA）分类器对左右手运动想象任务进行特征分类．最后，对不同特征的分类结果进行比较，得出以AAR＋累积频带能量作为特征在BCI系统中的优越性能．

Feature Extraction Method Based on AAR Model and Accumulated Band Power

A feature extraction method based on the combination of adaptive autoregressive （AAR） model parameters and accumulated band power was presented. The combination feature was used as feature vector to discriminate the left and right hand motor imagery in the brain-computer interface （BCI） system based on motor imagery. The perform- ance of BCI was improved through this method. Firstly, wavelet transformation and inverse transformation were adopted to decompose and reconstruct scalp electroencephalogram （EEG）. Noises in EEG data were filtered through this process. Different frequency band EEG signals were obtained. Secondly, the AAR model parameters and band power of different frequency bands were extracted. Then the feature extraction method based on accumulated band power feature and the combination of AAR with the accumulated band power were presented. With the AAR model parameters, the band power, the accumulated band power and the combination of AAR with the accumulated band power as feature vectors respectively; the linear discriminant analysis （LDA）classifier was used to discriminate left and right hands motor imagery tasks. Lastly, a comparison of classification results among the different features was conducted. The results show that the combined feature of AAR with the accumulated band power is superior to others in the BCI system.