摘要
脑机接口（Brain-computer Interface, BCI）是一种能在大脑与计算机或其他电子设备间建立直接通讯和控制通道的交互系统，在神经工程、康复和脑科学领域研究及应用越来越广泛。根据获取脑电信号方式的差异，BCI分为有创和无创。其中基于稳态视觉诱发电位（steady-state visual evoked potential, SSVEP）和运动想象的无创BCI是目前研究和应用最广泛的BCI系统。
虽然低频刺激模式下的SSVEP特征明显、易于检测。但低频刺激容易使受试者产生疲劳，甚至对某些受试者存在诱发光敏癫痫疾病的风险。因此，为提高舒适性和降低风险，采用高频刺激模式下的SSVEP是BCI研究的发展方向。然而，高频SSVEP特征不易检测，因此，如何在提高舒适度同时，保证特征分类效果，是基于SSVEP的BCI系统急需解决的问题。
另一方面，当受试者想象肢体运动时，在大脑感觉运动皮层区域，会出现事件相关同步和去同步现象（event-related desynchronization/synchronization,
ERD/ERS），这是运动想象EEG信号的主要特征。现有文献认为该现象主要反映在EEG信号幅值特征上，因此，对其在相位耦合特征上的表现，尤其是零相位耦合特征的意义，以及幅值和相位耦合特征之间的关系等，尚缺乏深入研究。目前运动想象BCI还存在训练时间长，涉及导联数较多等缺点。因此，如何选择导联，以及在少导联情况下，如何保证特征任务预测能力，对运动想象BCI研究具有重要意义。
为充分发挥高频刺激SSVEP和运动想象EEG的优点，克服低频刺激易疲劳和运动想象BCI盲现象等问题，通过对SSVEP和运动想象两种EEG信号特征的深入研究，本文提出少导联情况下，基于高频刺激SSVEP和运动想象EEG的新型混合BCI。此方法能有效降低单一模态BCI盲现象，并提高分类准确率。本文的主要内容包括以下几个方面：
①为提高SSVEP-BCI的舒适度，本文对比研究了低频和中高频SSVEP信号特征，提出了适合中高频SSVEP的最佳参考导联选择机制和特征提取方法，并将该方法应用于混合BCI的高频SSVEP特征分析中，从而在少导联情况下，有效提取高频SSVEP特征。
②通过对训练后与未训练运动想象相关EEG信号的幅值和相位耦合特征进行系统性分析，揭示了零相位耦合特征并非由容积导体效应导致的无效耦合。另外，通过采用经典溯源分析方法对运动想象EEG数据进行分析，结果表明经典溯源方法反映的是幅值特征的源，而相位耦合特征分析可以有效反映具有耦合关系
的源，验证了基于相位耦合脑电溯源分析是对经典溯源分析方法的补充。
③通过研究空间滤波对运动想象EEG幅值与相位耦合特征的影响，发现空间滤波能够将零相位耦合有
效信息加入到幅值特征中，使该特征得以增强。在此基础上，本文提出一种基于相位耦合的导联选择方法，为实现少导联、便携式BCI 提供方法支撑。与此同时，建立了一种融合相位耦合和幅值特征的新型特征提取方法，并与传统幅值和共同空间模式（common spatial pattern, CSP）特征提取方法进行对比分析。结果表明融合相位耦合的幅值特征提取方法在运动想象EEG信号特征分析中，优于传统的特征分析方法。
④结合①和③中的特征提取方法，提出了基于高频SSVEP和运动想象EEG 的新型混合BCI，受试者通过运动想象并注视高频刺激来控制光标向上或者向下运动。与单一模态的运动想象BCI和高频SSVEP-BCI的平均准确率进行对比可以发现，本文提出的混合BCI将平均准确率分别提高14%和2%，且有效解决了三位受试者的运动想象BCI盲问题。
综上所述，通过对两种常用EEG信号的特征分析，本文旨在提出一种提高用户舒适度，且具有更强普适性的混合BCI。基于Neuroscan采集系统和BCI2000软件平台，验证了新型混合BCI的有效性，同时也为后续BCI的基础研究提供了平台。与此同时，本文提出的关于BCI导联选择机制以及少导联BCI方法，对于可穿戴BCI设备开发，新型电极、电极帽的设计等，都具有重要指导意义。
关键词：稳态视觉诱发电位，运动想象，零相位耦合，空间滤波，混合BCI
ABSTRACT
Brain-computer interface (BCI) is an interaction system for the brain and the computer or other electrical equipments, which has become more and more popular in the area of neural engineering, rehabilitation and brain science. According to the different ways of brain signal acquisition, BCIs are generally divided into invasive and noninvasive BCI systems, in which the steady state visual evoked potential (SSVEP)- and motor imagery-bad noninvasive BCIs are most popular.
ps怎么复制选区The low-frequency stimulus evoked SSVEPs are easy to detect. However, the low frequency stimulus can easily cau fatigue and even the evoke of photonsitive epilepsy for tho who have the potential risk. Therefore, using high frequency stimulus to improve comfortability is the future work of SSVEP bad BCI study. However, the high-frequency stimulating SSVEP is hard to detect, which is the major concern.
美工区材料投放On the other hand, the event-related desynchronization and synchronization (ERD/ERS) over the nsorimotor cortex occurs when subjects imaging hands movement, which is the main feature of the motor imagery-bad EEG. According to the current documents, however, the ERD/ERS phenomenon is reflected in the amplitude feature. Thus, the pha-coupling feature for this phenomenon, the meaning of the zero-pha coupling and the relationship between amplitude and pha coupling are unclear. The current motor imagery BCI requires long-time training with a relative
打春牛
ly large number of channels involved. Therefore, how to lect channels and how to enhance the task prediction ability with less channels involved are also of vital importance in motor-imagery bad BCIs.
Given the advantages of the high-frequency SSVEP and the motor imagery bad EEG, this thesis provides a new hybrid BCI system according to the thorough feature analysis of SSVEP and motor imagery bad EEG. This hybrid BCI, which is bad on high frequency SSVEP and motor imagery bad EEG with less channels involved, overcomes the disadvantages of the low-frequency SSVEP and the BCI illiteracy phenomenon of motor-imagery bad BCI. It is able to decrea the BCI illiteracy effectively, and in the meantime increa the classification accuracy. The thesis is compod of the following four main aspects:
①To improve the comfortability of SSVEP-BCI, a comparison study of low-frequency and high-frequency SSVEP is carried out. And a method for choosing the best reference electrode and a feature extraction method are propod, which are
suitable for the medium and high frequency stimulus evoked SSVEP.
城市森林松木家具② According to the systematic study of the amplitude and pha-coupling feature of the trained and untrained motor-imagery related EEG, the zero-pha coupling feature is illustrated to be of important meaning, which cannot be explained by volume conduction. In addition, according to the analysis of motor-imagery bad EEG by the classical source localization method, the result shows that the classical source localization method reflects the amplitude-feature-bad source, but the pha-coupling method shows the sources of coupling relationship effectively. This verifies that pha-coupling method is a complement for the classical source location method.
③According to the study of the effects of the spatial filters on amplitude and pha-coupling features, it shows that spatial filtering is able to add the coupling feature into amplitude, and in the meantime enhance the amplitude feature. On this basis, this thesis propos a channel lection method bad on pha-coupling feature, which provide method-supports for the study of portable BCI using a small number of channels. In the meantime, a novel amplitude bad feature extraction method, involving pha coupling information, is propod. When comparing this method with the classical amplitude bad feature extraction and the common spatial filter (CSP) method, the result shows that this method is better than the classical feature extraction methods in motor imagery bad BCI.
古昭公路
④ A novel hybrid BCI involving high-frequency SSVEP and motor-imagery bad EEG is propod using the feature extraction methods provided in ①and ③. Subjects can move the cursor upward or downward by the imagery task and focusing on the high-frequency stimulus. This hybrid BCI can increa the classification accuracy by 14% and 2%, respectively, comparing with the pure motor-imagery or high-frequency SSVEP bad BCI.
To summarize, this thesis mainly propos the hybrid BCI with better comfortability and generosity according to the feature analysis of the two kinds of EEG signals. The novel hybrid BCI is validated successfully bad on the Neuroscan EEG acquisition system and BCI2000 software, which also provides a fundamental BCI rearch platform. In the meantime, the propod rule of how to choo electrodes for the BCI, especially with a small number of electrodes involved, is of great significance for the study of portable BCI and the design of new type of electrode and EEG acquisition cap.
Keywords: SSVEP, motor imagery, zero-pha coupling, spatial filter, hybrid BCI
目录
中文摘要.......................................................................................................................................... I 英文摘要....................................................................................................................................... III 1 绪
论.. (1)
1.1 BCI研究基础 (1)
1.1.1 BCI概念及现状介绍 (1)
1.1.2 BCI分类 (1)
1.2 EEG信号特征分析 (5)
1.2.1 一元变量分析方法 (5)
1.2.2 二元变量分析方法 (5)
1.2.3 多元变量分析方法 (6)
1.3 存在的问题和挑战 (8)
1.3.1 SSVEP-BCI舒适度 (8)
1.3.2 运动想象EEG零相位耦合与容积导体效应 (9)
梦见西红柿1.3.3 运动想象EEG零相位耦合与EEG溯源 (11)
1.3.4 空间滤波对运动想象BCI控制特征的影响 (11)
1.3.5 BCI盲现象及混合BCI (12)
1.4 本论文研究内容 (12)
2 低频与中高频SSVEP特征研究 (15)
2.1 SSVEP数据介绍 (15)
2.2 中高频和低频SSVEP特征比较研究 (17)
2.3 中高频SSVEP特征提取及分类算法研究 (22)
2.3.1 特征分类效果评估 (22)
2.3.2 中高频SSVEP最佳导联及频带特征计算 (22)
交通安全进校园2.3.3 中高频SSVEP分类 (25)
2.3.4 结果分析 (26)
高以翔个人资料简介2.4 本章小结 (27)
3 运动想象BCI零相位耦合特征研究 (29)
3.1 运动想象BCI零相位耦合特征 (29)
3.2 幅值与相位耦合特征提取 (30)
3.2.1 频率滤波 (30)
3.2.2 幅值特征提取 (30)
