基于EEG信号的认知任务模式分类研究

发布时间：2018-03-10 03:28

本文选题：运动想象分类　切入点：认知任务模式分类　出处：《杭州电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：认知科学是研究人类感觉、知觉、精神状态、大脑思维过程和信息处理过程的前沿性尖端学科，该方面的研究对揭示人脑之谜具有重要意义。认知任务的模式分类被广泛用于构建脑机交互系统、研究人脑的工作机制和各种脑疾病的发病机理。为了探索人脑的认知机理，本文重点针对两类认知任务-运动想象分类任务和驾驶疲劳状态分类任务基于脑电信号进行了深入研究。运动想象任务的模式分类是构建脑机交互系统的重要方式之一，构建该类系统的关键在于对执行不同肢体运动想象任务时的脑电进行特征提取，然后对提取的特征进行分类，并把特征分类结果转化为外部设备的控制命令。针对运动想象任务，本文重点研究了脑电信号的特征提取算法，主要做了两方面工作，一方面研究学习了多种经典特征提取算法，针对传统公共空间模式算法中滤波器成分选择方法的不足，提出了一种基于相关系数的新滤波器成分选择方法；另一方面本文根据传统微状态定义提出了一种广义微状态概念，并基于此广义概念提出了一种新的特征提取算法。本文使用国际BCI竞赛运动想象数据集和实验室自采集的数据集验证了上述两种算法的有效性。驾驶是一项涉及视觉、听觉、思维和判断等多种认知功能的复杂任务，如何区分长时间驾驶前后的警醒状态和疲劳状态是本文的另一个关注重点。为了达到研究目的，本文首先设计了一项模拟驾驶实验，搜集了长时间驾驶过程中的脑电数据；然后使用基于格兰杰因果关系构建的脑效应网络对比研究了驾驶员疲劳前后脑电信号模式的变化情况。该项研究发现了易受疲劳影响的大脑区域，并且发现脑效应网络的若干属性可以作为区分警醒状态和疲劳状态的指标。本文使用的研究方法一定程度上克服了当前多数疲劳检测算法不能衡量脑区之间的信息传递关系的缺点，研究结果对于实用疲劳检测系统的电极安放位置和检测指标的选择具有一定的指导意义。本文针对所选的两类认知任务进行了深入研究。在运动想象分类方面，针对传统公共空间模式算法提出的改进意见思路简洁，行之有效；基于广义微状态提取的特征包含了脑电模式的空间信息，分类简单。在疲劳状态检测和分类方面，基于脑效应网络展开研究，从网络角度可以全面衡量大脑活跃模式的全局特性和局部特性，，使用因果关系计算的效应连接可以在一定程度上反映不同脑区间的信息流向，方法新颖。两类任务具有很高的科研价值和实用意义。
[Abstract]:Cognitive science is a cutting-edge discipline that studies human feelings, perceptions, mental states, brain thought processes and information processing processes. The pattern classification of cognitive tasks is widely used to construct brain-computer interaction system, to study the working mechanism of human brain and the pathogenesis of various brain diseases, in order to explore the cognitive mechanism of human brain. This paper focuses on two kinds of cognitive tasks-motor imagination classification task and driving fatigue state classification task based on EEG. The pattern classification of motion imagination task is one of the important ways to construct brain-computer interaction system. The key of constructing this kind of system is to extract the feature of EEG when performing different limb motion imagination task, and then classify the extracted feature. The result of feature classification is transformed into the control command of external equipment. Aiming at the task of motion imagination, this paper focuses on the feature extraction algorithm of EEG signal, which is mainly done in two aspects. On the one hand, several classical feature extraction algorithms are studied, and a new filter component selection method based on correlation coefficient is proposed to solve the problem of filter component selection in the traditional common space pattern algorithm. On the other hand, based on the traditional definition of micro-state, a generalized concept of micro-state is proposed. Based on the generalized concept, a new feature extraction algorithm is proposed, and the validity of the above two algorithms is verified by using the image data set of international BCI competition and the self-collected data set of laboratory. Driving is a complex task involving many cognitive functions, such as vision, hearing, thinking and judgment. How to distinguish alertness and fatigue before and after driving for a long time is another focus of this paper. In this paper, a simulated driving experiment is designed to collect EEG data during a long driving period. Then the brain effect network based on Granger causality was used to compare the changes of EEG patterns in drivers before and after fatigue. The study found regions of the brain vulnerable to fatigue. It is also found that some attributes of the brain effect network can be used as indicators to distinguish alertness from fatigue. To some extent, the research method used in this paper overcomes the fact that most current fatigue detection algorithms can not measure the information between brain regions. The drawback of the information transfer relationship, The results can be used to guide the selection of electrode placement and detection index in practical fatigue detection system. In this paper, two kinds of cognitive tasks are studied in depth. In the classification of motion imagination, the improved ideas for the traditional public space model algorithm are simple and effective. The feature based on generalized micro-state extraction includes spatial information of EEG pattern, and the classification is simple. In the aspect of fatigue state detection and classification, the research is carried out based on brain effect network. The global and local characteristics of brain activity patterns can be comprehensively measured from a network perspective, and the effect connections calculated by causality can to some extent reflect the flow of information in different brain regions. The two kinds of tasks have high scientific research value and practical significance.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【参考文献】