共空间模式(CSP)广泛应用于脑电信号(EEG)的特征提取,合适的通道选择可以有效地提高CSP的分类性能,增加信噪比.根据运动想象信号的平均能量差来进行通道选择和特征提取.首先取两类运动想象信号的通道均值能量作为投票的阈值,根据投票差值统计各通道上有明显能量差值试次的数量,基于此来选择出合适的通道,然后对这些通道取能量特征进行归一化,再结合CSP空域特征利用SVM进行分类.在BCI Competition Ⅲ Data Sets Ⅳa和BCI Competition IV Dataset SetsⅠ两个数据集上进行的分类实验中,所提出的方法相比于全通道CSP,平均精度分别提高了5.7%和10.9%,通道数分别减少了74.3%和51.7%,验证了所提出的通道选择和特征提取方法的有效性.
Motor Imagery EEG Channel Selection and Feature Extraction Based on Average Energy Difference
Common spatial pattern(CSP)is widely used in the feature extraction of electroencephalogram(EEG).Appropriate channel se-lection can effectively improve the classification performance of CSP and increase the signal-to-noise ratio.Channel selection and feature extraction are performed according to the average energy difference of motor imagery signals.First,the channel mean energy of the two types of motor imagery signals is taken as the voting threshold,and the number of trials with obvious energy differences on each channel is counted according to the voting difference.The channel is normalized by energy features,and then combined with CSP airspace fea-tures to use SVM for classification.In the classification experiments on the BCI Competition Ⅲ Data Sets Ⅳa and BCI Competition Ⅳ Dataset Sets Ⅰ,compared with the full-channel CSP,the average accuracy of the proposed method is increased by 5.7％and 10.9％,and the number of channels is reduced by 74.3％and 51.7％,respectively,which verifies the effectiveness of the proposed channel selection and feature extraction method.
EEGmotor imageryCSPSVMchannel selectionenergy features
国家科技期刊平台
NSTL
万方数据
