融合眼电信息的多任务模式脑—机接口系统研究与实现

发布时间：2018-03-13 07:06

本文选题：混合脑-机接口　切入点：脑电信号　出处：《天津理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：混合型脑-机接口(Hybrid Brain Computer Interface,HBCI)是指在单一脑电信号模式基础上加入一种或多种新的控制方式,如另一种脑电信号模式或者是心率、眼动、血流变化等生理信号,进行混合控制的BCI。HBCI具有可完成任务种类多、识别准确率高、控制方式灵活等优点,已成为改进BCI系统性能研究的新方向。本文研究了融合运动想象脑电信号和眼动信号的混合脑-机接口在线控制系统,包括脑电和眼电信号的采集、预处理、特征提取和模式识别方法,并完成了实时控制小四轴飞行器的飞行操作实验。本文主要研究工作如下:(1)在分析比较了时域分析法、时频域分析法和共空间模式(Common Spatial Pattern,CSP)等特征提取方法的基础上,设计了OVO-CSP(One Versus One CSP)算法和OVR-CSP(One Versus the Rest CSP)算法提取多类别脑电信号的特征。(2)对EEG信号的模式识别方法进行了研究改进。详细分析了OVR-SVM(One Versus the Rest Support Vector Machine)、OVO-SVM(One Versus One Support Vector Machine)、DAG-SVM(DAG Support Vector Machine)等多种分类方法各自的优缺点,提出了一种新的双层SVM分类器结构对四类脑电信号进行分类。用2008年竞赛数据集data set 2a验证了该双层SVM分类器结构的分类正确率,并与DAG-SVM方法进行了对比。本文提出方法虽然训练时间比DAG-SVM略长,测试时间与DAG-SVM相差不多,而其分类正确率明显高于DAG-SVM。(3)提出用双树复小波变换(DTCWT)提取眼电信号的特征参数,解决了离散小波变换(DWT)提取眼电信号特征时存在的频率混叠、平移敏感性等缺陷。根据Donoho阈值去噪理论估计DTCWT分解后各层小波系数的阈值,阈值重构后进行下采样将其作为眼电信号的特征参数。采用SVM和时域特征相结合的方法进行分类,SVM分类器识别眼动为水平方向上的运动或垂直方向上的运动,再用眼电信号的时域特征识别水平或垂直方向上的眼动模式。实验结果显示,该分类方法能有效识别眼球的扫视模式和连续眨眼模式。(4)搭建了混合脑-机接口在线控制实验平台,在Emotiv EPOC脑电采集仪上扩展了眼电信号采集电极,将数据实时传回计算机。在MATLAB2015和VS2010上分别编写了GUI控制界面,设计了纠错机制。在线实验结果显示,该混合系统能够较好的完成圆点博士小四轴飞行器的起飞、前飞、左飞、右飞、后飞和降落等在线控制操作。
[Abstract]:Hybrid Brain Computer Interface (HBCI) refers to the addition of one or more new control methods to the single EEG mode, such as another EEG mode or physiological signals such as heart rate, eye movement, blood flow changes, etc. The BCI.HBCI with mixed control has many advantages, such as many kinds of tasks, high recognition accuracy, flexible control mode, and so on. It has become a new direction to improve the performance of BCI system. In this paper, a hybrid brain-computer interface control system is studied, which integrates motion imaginary EEG signal and eye movement signal, including EEG and Eye-signal acquisition and preprocessing. The method of feature extraction and pattern recognition, and the flight operation experiment of real-time control of small four-axis aircraft are completed. The main research work of this paper is as follows: 1) the time-domain analysis method is analyzed and compared. On the basis of time-frequency domain analysis and common Spatial patterns, The OVO-CSP(One Versus One CSP algorithm and OVR-CSP(One Versus the Rest CSP algorithm are designed to extract the features of multi-class EEG signals. (2) the pattern recognition method of EEG signal is studied and improved. The OVR-SVM(One Versus the Rest Support Vector Vector machine machine is analyzed in detail, and the OVR-SVM(One Versus the Rest Support Support Vector Vector Versus One One Support Support Vector machine machine is analyzed in detail. The advantages and disadvantages of various classification methods, A new two-layer SVM classifier structure is proposed to classify four kinds of EEG signals. The classification accuracy of the two-layer SVM classifier structure is verified by data set 2a of 2008 contest data set. In this paper, though the training time is a little longer than that of DAG-SVM, the test time is not different from that of DAG-SVM, and the classification accuracy is obviously higher than that of DAG-SVM.TWT3. The defects of frequency aliasing and translational sensitivity in extracting Eye-signal features by discrete wavelet transform (DWT) are solved. According to the Donoho threshold denoising theory, the threshold of wavelet coefficients in each layer after DTCWT decomposition is estimated. The method of combining SVM and time domain features is used to classify the motion of eye movement in horizontal direction or vertical direction. Then the time domain feature of Eye-signal is used to identify the eye movement pattern in horizontal or vertical direction. The experimental results show that, This classification method can effectively identify the scanning mode of eyeball and the continuous blinking mode. (4) A mixed brain-computer interface on-line control experimental platform is built, and the Eye-electric signal collecting electrode is extended on the Emotiv EPOC EEG acquisition instrument. The data is transmitted back to the computer in real time. The GUI control interface is written on MATLAB2015 and VS2010, and the error correction mechanism is designed. The online experimental results show that the hybrid system can accomplish the take-off, forward flight and left flight of the small four-axis aircraft. Right flight, rear flight and landing and other online control operations.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【相似文献】