摘要
为提高运动想象脑机接口的分类正确率,结合小波包分解与近似熵对脑电信号进行特征提取。该方法利用小波包对脑电信号全频段进行分解,用近似熵函数对小波包结点提取分类特征,然后用稀疏表示对特征向量进行降维,最后使用功率差方法进行分类。实验结果表明,在使用1秒数据进行分类的条件下,该方法在使用2种不同通道集合时都取得了很好的分类效果。使用32个和10个通道时分类正确率分别达到了95.65%和86.41%,比小波包分解与空域滤波方法分别提高了5.9%和8.32%,比传统的共空域模式方法分别提高了7.18%和7.27%。另外,使用的数据长度越短,分类识别率越高,表明该方法更适用于较短的数据,有利于提高脑机接口的信息传输速度。
To improve the classification accuracy of motor imagery-based brain-computer interface（BCI）systems,a novel method for EEG feature extraction based on wavelet packet decomposition and approximate entropy is proposed.The method uses wavelet packet to decompose the whole frequency band of EEG signals and uses approximate entropy to compute the feature values for the obtained wavelet packet nodes.The feature vectors are dimensionally reduced by sparse representation and finally are classified by apower difference method.Experimental results showed that in the case of EEG data of length 1s being used,the method achieved high classification accuracies of 95.65% and 86.41% with 32 and 10 channels respectively in two-class motor imagery based BCIs,which are 5.9% and 8.32% higher than those achieved by the combined method of wavelet transform and the spatial filtering respectively,and 7.18% and 7.27% higher than the traditional methods of spatial pattern（CSP）respectively.Moreover,the proposed method is more suitable for BCI application with shorter data length,which is helpful for improving information transfer speed in motor imagery based BCIs.
出处
《南昌大学学报（理科版）》
CAS
北大核心
2017年第3期282-287,共6页
Journal of Nanchang University(Natural Science)
基金
国家自然科学基金项目(61365013
61663025)
江西省教育厅科技项目(GJJ13054)
关键词
脑机接口
运动想象
小波包分解
近似熵
稀疏表示
共空域模式
brain-computer interface
motor imagery
wavelet packet decomposition
approximate entropy
sparse representation
common spatial pattern
