基于多特征融合的运动想象脑电信号分类研究
|
摘要
针对脑机接口(BCI)系统中常用的运动想象脑电信号,提出多特征融合的方法,提高其分类准确率。采用自回归(AR)模型、小波包变换、共空间模式(CSP)对BCI2003竞赛数据进行信号处理,提取AR模型系数、小波包熵差、方差作为特征,并在分类器上测试准确率。实验结果表明:多特征融合的分类准确率均高于单特征,保持在90%左右,可以更好地表征运动想象脑电信号。数据处理所选取的时间段在4-7s,实验结果具有较好的鲁棒性和泛化能力。
Proposes a method of multi-features fusion to improve the classification accuracy of motor imagery electroencephalogram signals commonly used in brain computer interface(BCI) system. Uses the autoregressive(AR) model, wavelet packet transform and common spatial pattern(CSP) to process BCI2003 competition data. AR model coefficient, extracts the wavelet package entropy difference and variance as the characteristic, and tests the accuracy on the classifier. The experimental results show that the classification accuracy of the multi-features fusion is higher than that of the single feature. It is maintained at about 90%, which can represent the motor imagery electroencephalogram signals better than the single feature. The time period of data processing is 4-7 s, and the experimental results have good robustness and generalization ability.
Proposes a method of multi-features fusion to improve the classification accuracy of motor imagery electroencephalogram signals commonly used in brain computer interface(BCI) system. Uses the autoregressive(AR) model, wavelet packet transform and common spatial pattern(CSP) to process BCI2003 competition data. AR model coefficient, extracts the wavelet package entropy difference and variance as the characteristic, and tests the accuracy on the classifier. The experimental results show that the classification accuracy of the multi-features fusion is higher than that of the single feature. It is maintained at about 90%, which can represent the motor imagery electroencephalogram signals better than the single feature. The time period of data processing is 4-7 s, and the experimental results have good robustness and generalization ability.
引文
[1]Rajesh P. N. Rao. Brain-Computer Interfacing:An Introduction[M].张莉,陈民铀,北京:机械工业出版社,2016:1-3.
[2]耿雪青,佘青山,韩笑,等.基于人工蜂群优化高斯过程的运动想象脑电信号分类[J].传感技术学报,2017,30(3),379-384.
[3]王卫星,孙守迁,李超,等.基于卷积神经网络的脑电信号上肢运动意图识别[J].浙江大学学报(工学版),2017,51(7):1382-1389.
[4]马也,常天庆,郭理彬.运动想象脑电信号特征提取与分类算法研究[J].计算机工程与应用,2017,53(16):149-154.
[5]王艳景,乔晓艳,李鹏.基于小波包熵和支持向量机的运动想象任务分类研究[J].仪器仪表学报,2010,31(12):2730-2735.
[6]徐宝国,宋爱国,王爱民.基于小波包能量的脑电信号特征提取方法[J].东南大学学报(自然科学版),2010,40(6):1204-1206.
[7]郭红想,严军,王典洪,等.基于小波包和组合分类器的脑电信号分类[J].计算机工程与应用,2016,52(18):148-153.
[8]黄思娟,吴效明.基于能量特征的脑电信号特征提取与分类[J].传感技术学报,2010,23(6):783-785.
[9]赵立军,朱晓军.基于局部值分解与样本熵的脑电信号特征提取与分类[J].计算机工程,2017,43(2):299-303.
[10]李明爱,崔燕,杨金福,等.基于HHT和CSSD的多域融合自适应脑电特征提取方法[J].电子学报,2013,41(12):2480-2486.
[11]王力,张雄.小波滤波与AR模型在脑电信号处理的应用[J].电子器件,2012,35(4):462-464.
[12]张毅,罗明伟,罗元.脑电信号的小波变化和样本熵特征提取方法[J].智能系统学报,2012,7(4):340-344.
[13]http://bbci.de/cometition/ii/download.
[14]罗志增,周镇定,周英,等.双树复小波特征在运动想象脑电识别中的应用[J].传感技术学报,2014,27(5):576-580.
[15]Ramoser H,Muller-Gerking J,Pfurtscheller G. Optimal Spatial Filtering of Single Trial EEG during Imagined Hand Movement[J]. IEEE Trans. on Rehab. 2000,8(4):441-46.
[2]耿雪青,佘青山,韩笑,等.基于人工蜂群优化高斯过程的运动想象脑电信号分类[J].传感技术学报,2017,30(3),379-384.
[3]王卫星,孙守迁,李超,等.基于卷积神经网络的脑电信号上肢运动意图识别[J].浙江大学学报(工学版),2017,51(7):1382-1389.
[4]马也,常天庆,郭理彬.运动想象脑电信号特征提取与分类算法研究[J].计算机工程与应用,2017,53(16):149-154.
[5]王艳景,乔晓艳,李鹏.基于小波包熵和支持向量机的运动想象任务分类研究[J].仪器仪表学报,2010,31(12):2730-2735.
[6]徐宝国,宋爱国,王爱民.基于小波包能量的脑电信号特征提取方法[J].东南大学学报(自然科学版),2010,40(6):1204-1206.
[7]郭红想,严军,王典洪,等.基于小波包和组合分类器的脑电信号分类[J].计算机工程与应用,2016,52(18):148-153.
[8]黄思娟,吴效明.基于能量特征的脑电信号特征提取与分类[J].传感技术学报,2010,23(6):783-785.
[9]赵立军,朱晓军.基于局部值分解与样本熵的脑电信号特征提取与分类[J].计算机工程,2017,43(2):299-303.
[10]李明爱,崔燕,杨金福,等.基于HHT和CSSD的多域融合自适应脑电特征提取方法[J].电子学报,2013,41(12):2480-2486.
[11]王力,张雄.小波滤波与AR模型在脑电信号处理的应用[J].电子器件,2012,35(4):462-464.
[12]张毅,罗明伟,罗元.脑电信号的小波变化和样本熵特征提取方法[J].智能系统学报,2012,7(4):340-344.
[13]http://bbci.de/cometition/ii/download.
[14]罗志增,周镇定,周英,等.双树复小波特征在运动想象脑电识别中的应用[J].传感技术学报,2014,27(5):576-580.
[15]Ramoser H,Muller-Gerking J,Pfurtscheller G. Optimal Spatial Filtering of Single Trial EEG during Imagined Hand Movement[J]. IEEE Trans. on Rehab. 2000,8(4):441-46.