基于卡尔曼滤波脑认知动态特征提取方法研究

发布时间：2018-01-22 11:25

本文关键词： 脑电信号卡尔曼滤波动态分析特征提取单次诱发成份脑网络　出处：《哈尔滨工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：现今,在脑科学和信号处理技术的迅猛发展下,脑机接口设备逐渐应用于临床医学、军事、游戏等领域。脑认知特征提取方法是脑机接口设备研发基础,由于脑机接口技术需要实时地完成人机交互任务,所以脑认知特征提取方法也急需一种具有高时间分辨率和能够实时分析的脑电信号处理方法。本文利用Kalman滤波方法进行认知活动特征提取。将脑认知特征提取问题划分成两个子问题:单通道脑电信号特征提取和多通道脑电信号特征提取。(1)在单通道脑电信号研究部分,本文提出了单次诱发成份的自适应卡尔曼滤波提取方法,以及自适应诱发成份优势区间分离机制算法。该算法利用了脑电诱发成份和自发成份的幅值变化速率的差异,定义了测量余量和估计幅值两个参数,以两参数为标准该算法可自动区分脑电信号中诱发成份优势区间和噪声优势区间,实现单通道脑电信号单一或少量诱发成份的精确提取。同时进行了针对模拟诱发成份和真实诱发成份的方法验证实验,其中真实诱发成份数据通过Oddball范式下的视觉脑认知实验获得。并将该方法应用于脑机接口系统中的单次诱发成份检测,结果表明该方法可有效提升脑机接口系统的识别正确率,高达92%。(2)在多通道脑电信号研究部分,本文建立语音冲突认知活动模型,提出基于Kalman滤波状态追踪的时变脑网络分析方法。该方法在现有的多变量自回归模型的基础上,采用Kalman滤波方改进多变量自回归模型的系数估计,再通过有向传递函数的因果分析方法获取包含频域信息的效应脑网络模型。同时以Stroop实验范式采集了听觉脑认知实验数据,进行了语音冲突脑网络构建与分析实验,结果表明该方法能够提取了语音冲突的脑网络的结构特征和频域特征,解决了脑认知活动下大脑内部信息传递的获取,以及大脑神经连接动态演化过程的探究。本文的所提出的认知活动动态特征提取方法能够很好地分析脑电信号时域、频域、空间域信息,为脑电信号建模和特征提取提供了新思路。
[Abstract]:Nowadays, with the rapid development of brain science and signal processing technology, brain-computer interface equipment is gradually applied in clinical medicine, military, games and other fields. Brain cognitive feature extraction method is the basis of brain-computer interface equipment research and development. Because the brain-computer interface technology needs to complete the human-computer interaction task in real time. Therefore, a method of EEG processing with high temporal resolution and real time analysis is urgently needed. In this paper, we use Kalman filter to extract the feature of cognitive activity. The feature extraction problem is divided into two sub-problems:. Single channel EEG feature extraction and multichannel EEG feature extraction. (. 1) in the single channel EEG research section. In this paper, an adaptive Kalman filter extraction method for single evoked components is proposed. This algorithm takes advantage of the difference of amplitude change rate between EEG evoked component and spontaneous component, and defines two parameters, measurement margin and estimation amplitude. Using two parameters as the standard, the algorithm can automatically distinguish the dominant region of evoked component from the dominant interval of noise in EEG signal. The single or small amount of evoked components of single channel EEG signals can be extracted accurately. At the same time, the methods of simulating evoked components and real evoked components are verified. The real evoked component data were obtained by the visual brain cognitive experiment under Oddball paradigm, and the method was applied to the single evoked component detection in the BCI system. The results show that this method can effectively improve the recognition accuracy of BCI system, up to 92 / 2) in the multi-channel EEG research part, this paper establishes a speech conflict cognitive activity model. A time-varying brain network analysis method based on Kalman filter state tracking is proposed, which is based on the existing multivariate autoregressive model. The Kalman filter is used to improve the coefficient estimation of multivariate autoregressive model. Then the effect brain network model containing frequency domain information was obtained by causality analysis of directed transfer function and the auditory brain cognitive experiment data were collected by Stroop experimental paradigm. The experimental results show that this method can extract the structural features and frequency domain features of the brain network of speech conflict and solve the problem of information transmission in the brain under the brain cognitive activity. As well as the exploration of the dynamic evolution process of cerebral nerve connection. The proposed method of dynamic feature extraction of cognitive activity can well analyze the EEG information in time domain, frequency domain and spatial domain. It provides a new idea for EEG modeling and feature extraction.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【相似文献】