【作者】 周旭；

【导师】 马玉良；

【作者基本信息】 杭州电子科技大学 ， 控制工程， 2017， 硕士

【摘要】 脑机接口技术不依赖于传统的神经肌肉系统,可实现与外界的通讯与交流。本文的研究目的是将脑机接口应用到康复工程中康复机器人的控制及运动想象的反馈中,提高康复训练效果。针对于实际应用中遇到的信号类型单一、识别率低、患者与健康人之间的特征差异等关键技术问题,本文展开了具体研究。主要研究成果包括:1)本文提出了基于EEG与fNIRS的多模态脑机接口,根据两种信号时间分辨率的不同设计了新的实验范式。采集到同步的多模态信号后,为消除EEG信号中的噪音,先对EEMD分解后的高频分量进行小波阈值处理,再与低频部分进行重构得到消噪后的信号。通过两种方法结合起来运用,优势互补,既能够获得更明显的消噪效果,又能最大程度地保留信号的有效信息。2)根据脑电信号具有明显的非线性特点,采用非线性动力学分析方法提取特征,同时考虑到通道之间的相互联系,又提取了信号的CSP特征,结果表明在多通道的情况下CSP的分类效果更好。提取多模态信号的特征后,再利用BP神经网络算法对其进行分类,为避免神经网络的初始权值、阈值的随机性对分类效果的影响,提出GA优化BP神经网络,实验结果证明了优化算法的有效性。3)本文的研究目的是将脑机接口应用到运动功能的康复训练中。基于此想法,本文对健康人和脑卒中患者运动想象时的多模态信号进行比较分析,结果发现两者之间并没有显著性差异。所以,基于健康人的脑机接口也适用于脑卒中患者。并且,实验结果表明两者的特征之间具有一定的互补性,所以用于患者的脑机接口也应加入健康受试者的信息。更多还原

【Abstract】 Brain-Computer Interface(BCI)technique enables the user to communicate and control external devices by using electrical signals detected on the scalp since it does not depend on conventional neuromuscular control.In this work,BCI is used as an effective tool to actively control rehabilitation robot as well as evaluate the performance of motor imagery(MI)to improve the functional recovery after the rehabilitation training.Several technical problems have been considered and taken into account of most-influenced factors in current clinical applications of BCI: limited signal types,low target identification accuracy as well as characteristic difference between dysfunctional patients and normal people.The main research results include:Firstly,based on the temporal resolution of EEG and fNIRS,a EEG-fNIRS hybrid BCI technique is proposed and exploded to design new experimental task.And a denoising technique based on the Ensemble Empirical mode decomposition(EEMD)is put forward to analyze the multi-modal signals simultaneously collected by combined EEG&fNIRS devices.That is,the denoised signals can be reconstructed on the basis of low frequency information and high frequency information which has been filtered out via wavelet threshold denoising method.It has been demonstrated that more useful signals have been well kept after noise reduction due to the complementary signals obtained from EEMD and wavelet threshold denoising method.Secondly,since EEG signals are severely nonlinear,a nonlinear dynamics analysis approach has been used to characterize EEG signals in terms of spatial patterns.Also taking into account of the interactions between each channel,common spatial pattern(CSP)method is employed for classification and multiple channels have been validated to provide high classification accuracies.To avoid the influence of initial value and random threshold determination strategy on the back-propagation(BP)neural networks classification algorithm,GA optimization method is applied and numerous experiments are performed to show the efficiency of the proposed algorithm.Finally,multimodality based BCI technique is applied to the rehabilitation training of disabled patients.It has been demonstrated that no obvious difference is observed between normal subjects and dysfunctional patients in the multimodal signals during the MI process.Therefore,in consideration of the complementary property between signals from normal subjects and patients,the information of normal subjects should be added as supplementary information in the design of BCI system.更多还原