运动观察脑电特征分析与识别

发布时间：2018-03-26 12:10

本文选题：运动观察　切入点：脑电信号　出处：《郑州大学》2017年硕士论文

【摘要】：运动观察作为人脑的一种认知活动,对运动观察过程中脑电(Electroencephalogram,EEG)信号的研究,有利于对人脑工作机制的探索。而且通过对不同运动观察过程EEG的特征提取与识别,在军事侦查、目标追踪中也有很大的应用价值,也为脑-机接口系统设计提供一种新的思路。然而在运动观察过程中,由于大脑没有主动的思维任务参与,无法通过EEG直接确定是否处于有效运动观察状态,且与运动想象、稳态视觉诱发电位相比,运动观察过程的脑电信号幅值更弱,更加难以获取。本文以实现观察小车左转、右转过程的脑电信号特征解析与识别为目的,首先采用SMI眼动仪和Neuroscan脑电设备同步采集信号,设计了观察小车左转、右转两种状态的实验范式,利用眼动轨迹信号分析来确定有效运动观察任务。从时频角度分析有效运动观察过程的激活脑区和不同频段能量谱分布,确定特征明显频段。由于人脑在认知活动中,其神经元之间存在有向信息的交互,进一步采用能够描述不同脑区间信息流向的因果网络分析方法,通过分析运动观察过程中因果网络的网络测度,找到其差异性,并对差异性明显的网络测度进行分类。最后,利用CSP和SVM算法对运动观察EEG特征进行识别。主要研究内容如下:(1)针对运动观察EEG高度非平稳低信噪比的特点,以研究获取运动观察过程中脑电特征明显频段作为切入点,对有效任务的EEG进行预处理,提高EEG信噪比;然后,对不同频段EEG进行脑地形图分析,定位激活脑区、确定关键通道;最后,运用WPT和功率谱融合的方法,分析关键通道EEG在不同频段范围内的能量谱分布,确定特征明显频段。结果显示:特征明显频段为0.49-0.98Hz。(2)基于不同脑区间存在信息流,采用因果网络测度差异性分析的方法研究运动观察信号特征,利用GC、DTF、PDC三种分析方法对不同频段EEG进行因果网络构建。通过分析不同阈值下因果网络的网络密度和全局效率,选择合适的阈值,并分析网络测度(包括度、聚类系数、全局效率)的差异性。结果表明,在0-4Hz上,GC值的聚类系数具有显著性差异。(3)针对运动观察过程中脑电特征的识别问题,利用CSP算法对EEG进行滤波,以滤波后的信号能量为特征,并采用SVM进行特征识别,比较了不同频段上EEG的分类识别率,最高为0-4Hz上的86.15%。最后,通过对通道进行优化,可以在较少通道的情况下实现较高的分类准确率,并实现了基于GC的聚类系数特征的分类识别。
[Abstract]:As a cognitive activity of human brain, the study of electroencephalograms (EGG) signals in the course of motion observation is beneficial to the exploration of the working mechanism of human brain. Moreover, by extracting and recognizing the characteristics of EEG in different motion observation processes, it can be used in military investigation. Target tracking also has great application value and provides a new way of thinking for the design of brain-computer interface system. However, in the process of motion observation, the brain has no active thinking task to participate. It is impossible to determine directly by EEG whether it is in an effective state of motion observation, and the amplitude of EEG during exercise observation is weaker and more difficult to obtain than that of motion imagination. For the purpose of analyzing and recognizing the characteristics of EEG signals in the process of right turn, the SMI eye movement instrument and the Neuroscan EEG equipment are used to synchronize the acquisition of signals, and an experimental paradigm is designed to observe the two states of the vehicle turning left and right. The effective motion observation task is determined by using eye movement track signal analysis. The active brain region and the energy spectrum distribution of different frequency bands in the effective motion observation process are analyzed from the angle of time and frequency, and the characteristic obvious frequency band is determined. Because the human brain is in the cognitive activity, There is the interaction of directed information between neurons. The causal network analysis method, which can describe the flow of information in different brain regions, is further used to find the difference of the causal network through analyzing the network measurement of the causal network in the course of motion observation. Finally, CSP and SVM algorithms are used to recognize the feature of motion observation EEG. The main research contents are as follows: 1) aiming at the characteristics of high non-stationary and low signal-to-noise ratio (SNR) of motion observation EEG, In order to improve the signal-to-noise ratio (SNR) of EEG, the EEG of effective task is pretreated with the obvious frequency band of EEG characteristics in the course of motion observation. Then, the brain topographic map of EEG in different frequency bands is analyzed to locate and activate the brain region. Finally, using the method of WPT and power spectrum fusion, the energy spectrum distribution of critical channel EEG in different frequency range is analyzed. The results show that the obvious characteristic frequency range is 0.49-0.98Hz.z.t2) based on the existence of information flow in different brain regions, the motion observation signal features are studied by using the method of difference analysis of causal network measure. The causality network of EEG in different frequency bands is constructed by using three analysis methods. By analyzing the network density and global efficiency of causality network under different thresholds, the appropriate threshold is selected, and the network measure (including degree, clustering coefficient) is analyzed. The results show that the clustering coefficients of GC values on 0-4Hz have significant difference. Aiming at the problem of EEG feature recognition during motion observation, the EEG is filtered by CSP algorithm, which is characterized by the filtered signal energy. The recognition rate of EEG in different frequency bands is compared with that of 86.15 on 0-4Hz. Finally, by optimizing the channel, the classification accuracy can be achieved in the case of fewer channels. The classification and recognition of clustering coefficient features based on GC are realized.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【参考文献】