多类运动想象脑电信号的两级特征提取方法

共同空间模式（Common spatial pattern，CSP）是运动想象脑机接口（Brain-computer interface，BCI）中常用的特征提取方法，但对多类任务的分类正确率却明显低于两类任务.通过引入堆叠降噪自动编码器（Stacked denoising autoencoders，SDA），提出了一种多类运动想象脑电信号（Electroencephalogram，EEG）的两级特征提取方法.首先利用一对多CSP（One versus rest CSP，OVR-CSP）将脑电信号变换到使信号方差区别最大的低维空间，然后通过SDA网络提取其中可以更好表达类别属性的高层抽象特征，最后使用Softmax分类器进行分类.在对BCI竞赛IV中Data-sets 2a的4类运动想象任务进行的分类实验中，平均Kappa系数达到0.69，表明了所提出的特征提取方法的有效性和鲁棒性.

Two-level Feature Extraction Method for Multi-class Motor Imagery EEG

Common spatial pattern (CSP) is a popular method of feature extraction for motor imagery based brain-computer interface (BCI). However, the classification accuracy of multi-class tasks is obviously lower than that of two-class tasks with CSP. By employing the stacked denoising autoencoders (SDA), a two-level feature extraction method for multi-class motor imagery electroencephalogram (EEG) is proposed. Firstly, one versus rest CSP (OVR-CSP) is adopted to convert EEG into low dimensional space in which the discrimination of signal variances is maximized. Then, SDA network is used to extract the higher level abstract features which can characterize the category attributes more effectively. Finally, the motor imagery tasks are classified with Softmax classifier. In the classification experiment with four-class motor imagery tasks from Data-sets 2a of the BCI competition IV, this method achieves the average Kappa value of 0.69. The results show that the proposed method is effective and robust.