EMD融合PSD、CSP的脑电特征提取方法
|
摘要
为了提高运动想象分类精确度,提出一种基于经验模式分解(EMD),并结合功率谱密度(PSD)和公共空间模式(CSP)的特征提取算法。首先将采集的脑电信号进行预处理,再对信号使用EMD算法得到多个固有模态函数(IMFs)。通过计算每次实验原始脑电信号与各阶IMF分量之间的相关系数,并计算所有实验得出的相关系数的绝对值的平均数,选择具有较大相关系数绝对值平均数的固有模态函数,计算其功率谱密度作为特征,经共空间模式投影映射再提取相应的特征向量,并用支持向量机(SVM)进行分类。对9名受试者的运动想象进行分类结果分析,得到的平均分类正确率在96%以上。最后将该方法与其他方法做比较,证明了该算法的可行性。
In order to improve the classification accuracy of motor imagery,we propose a feature extraction algorithm based on empirical mode decomposition(EMD) combined with power spectral density(PSD) and common space pattern(CSP). First,the collected EEG signal is preprocessed,and then EMD algorithm is used to obtain multiple natural modal functions(IMFs). The correlation coefficient between the original EEG signal and the IMF components are calculated,and the average of the absolute values of the correlation coefficients derived from all experiments is computed. The intrinsic modal function with an average absolute number of large correlation coefficients is selected,and its power spectral density is calculated as a feature. The corresponding feature vector is extracted by the common space pattern projection mapping and classified by a support vector machine(SVM). The classification of the motor imagery of 9 subjects is analyzed,and the average classification accuracy is above 96%. Finally,this method is compared with other methods to prove its the feasibility.
In order to improve the classification accuracy of motor imagery,we propose a feature extraction algorithm based on empirical mode decomposition(EMD) combined with power spectral density(PSD) and common space pattern(CSP). First,the collected EEG signal is preprocessed,and then EMD algorithm is used to obtain multiple natural modal functions(IMFs). The correlation coefficient between the original EEG signal and the IMF components are calculated,and the average of the absolute values of the correlation coefficients derived from all experiments is computed. The intrinsic modal function with an average absolute number of large correlation coefficients is selected,and its power spectral density is calculated as a feature. The corresponding feature vector is extracted by the common space pattern projection mapping and classified by a support vector machine(SVM). The classification of the motor imagery of 9 subjects is analyzed,and the average classification accuracy is above 96%. Finally,this method is compared with other methods to prove its the feasibility.
引文
[1] 孙进,张征,周宏甫.基于脑机接口技术的康复机器人综述[J].机电工程技术,2010,39(4):13-16.
[2] YANG Banghua,LI Huarong,WANG Qian,et al.Subject-based feature extraction by using fisher WPD-CSP in brain-computer interfaces[J].Computer Methods & Programs in Biomedicine,2016,129:21-28.
[3] 徐晓刚,徐冠雷,王孝通,等.经验模式分解(EMD)及其应用[J].电子学报,2009,37(3):581-585.
[4] 刘锦,吴小培,周蚌艳,等.单次样本对的CSP滤波器设计及其在脑电训练样本优化中的应用[J].信号处理,2017,33(7):993-1001.
[5] 李明爱,刘净瑜,郝冬梅.基于改进CSP算法的运动想象脑电信号识别方法[J].中国生物医学工程学报,2009,28(2):161-165.
[6] 叶竞,石锐,何庆华.基于HHT和改进CSP算法的运动想象BCI系统[J].重庆理工大学学报:自然科学版,2012,26(5):70-73.
[7] 陈悦,张少白.LM算法在神经网络脑电信号分类中的研究[J].计算机技术与发展,2013,23(2):119-122.
[8] 罗志增,李亚飞,孟明,等.脑电信号的混沌分析和小波包变换特征提取算法[J].仪器仪表学报,2011,32(1):33-39.
[9] 王登,苗夺谦,王睿智.一种新的基于小波包分解的EEG特征抽取与识别方法研究[J].电子学报,2013,41(1):193-198.
[10] 于路,薄华.基于改进EMD 的运动想象脑电信号识别算法研究[J].微型机与应用,2016,35(9):58-61.
[11] GUPTA S S,SOMAN S,RAJ P G,et al.Improved classification of motor imagery datasets for BCI by using approximate entropy and WOSF features[C]//International conference on signal processing and integrated networks.Noida,India:IEEE,2014:90-94.
[12] LEEB R,LEE F,KEINRATH C,et al.Brain-computer communication:motivation,aim,and impact of exploring a virtual apartment[J].IEEE Transactions on Neural Systems & Rehabilitation Engineering,2007,15(4):473-482.
[13] ROBINSON N,VINOD A P,KAI K A,et al.EEG-based classification of fast and slow hand movements using wavelet-CSP algorithm[J].IEEE Transactions on Biomedical Engineering,2013,60(8):2123-2132.
[14] XU P,DAVOINE F,ZHA H,et al.Evidential calibration of binary SVM classifiers[J].International Journal of Approximate Reasoning,2016,72:55-70.
[2] YANG Banghua,LI Huarong,WANG Qian,et al.Subject-based feature extraction by using fisher WPD-CSP in brain-computer interfaces[J].Computer Methods & Programs in Biomedicine,2016,129:21-28.
[3] 徐晓刚,徐冠雷,王孝通,等.经验模式分解(EMD)及其应用[J].电子学报,2009,37(3):581-585.
[4] 刘锦,吴小培,周蚌艳,等.单次样本对的CSP滤波器设计及其在脑电训练样本优化中的应用[J].信号处理,2017,33(7):993-1001.
[5] 李明爱,刘净瑜,郝冬梅.基于改进CSP算法的运动想象脑电信号识别方法[J].中国生物医学工程学报,2009,28(2):161-165.
[6] 叶竞,石锐,何庆华.基于HHT和改进CSP算法的运动想象BCI系统[J].重庆理工大学学报:自然科学版,2012,26(5):70-73.
[7] 陈悦,张少白.LM算法在神经网络脑电信号分类中的研究[J].计算机技术与发展,2013,23(2):119-122.
[8] 罗志增,李亚飞,孟明,等.脑电信号的混沌分析和小波包变换特征提取算法[J].仪器仪表学报,2011,32(1):33-39.
[9] 王登,苗夺谦,王睿智.一种新的基于小波包分解的EEG特征抽取与识别方法研究[J].电子学报,2013,41(1):193-198.
[10] 于路,薄华.基于改进EMD 的运动想象脑电信号识别算法研究[J].微型机与应用,2016,35(9):58-61.
[11] GUPTA S S,SOMAN S,RAJ P G,et al.Improved classification of motor imagery datasets for BCI by using approximate entropy and WOSF features[C]//International conference on signal processing and integrated networks.Noida,India:IEEE,2014:90-94.
[12] LEEB R,LEE F,KEINRATH C,et al.Brain-computer communication:motivation,aim,and impact of exploring a virtual apartment[J].IEEE Transactions on Neural Systems & Rehabilitation Engineering,2007,15(4):473-482.
[13] ROBINSON N,VINOD A P,KAI K A,et al.EEG-based classification of fast and slow hand movements using wavelet-CSP algorithm[J].IEEE Transactions on Biomedical Engineering,2013,60(8):2123-2132.
[14] XU P,DAVOINE F,ZHA H,et al.Evidential calibration of binary SVM classifiers[J].International Journal of Approximate Reasoning,2016,72:55-70.