基于AR模型和SVM的脑电信号分类

 基于P300事件相关电位的脑机接口（BCI）系统中，有效的P300特征提取及分类是系统开展后续工作的关键。应用时间序列自回归（AR）模型及支持向量机（SVM）算法对脑电信号进行P300分类；对10导联脑电数据分别分段，并对每段建立AR模型；采用最小二乘法进行AR模型系数估计，由估计出的系数序列构成特征向量，送入SVM进行模式分类。实验针对BCI Competition Ⅲ dataset Ⅱ数据集进行了方法验证，提出的方法在15试次情况下识别正确率达93.5%。实验及数据分析结果表明，应用SVM分类器对AR模型提取出的系数序列特征向量进行分类，具有较好的系统识别正确率，可为实现基于P300的BCI系统实际应用奠定理论和实验基础。

EEG Signals Classification Based on AR Model and SVM Algorithm

In P300 based brain-computer interface (BCI), the effective feature exaction and classification of P300 is the key to carry out the follow-up work. An electroencephalogram (EEG) classification method combining with autoregressive (AR) model and support vector machine (SVM) was proposed. For 10 channels EEG data, AR model was built up for each epoch. The estimation of AR coefficients was taken on using least square method and the estimated coefficient sequences constituted the feature vectors. SVM was used as classifier and dataset Ⅱ of BCI Competition Ⅲ was used to verify this method. The recognition accuracy arrived at 93.5% with 15 times stimulations. The experimental results and data analysis show that the method using SVM to classify the feature vectors composed of AR coefficient sequences owns satisfactory recognition accuracy. It lays good comparison theory and experimental basis for the realization of P300 based BCI.