基于时频分析的头皮脑电特征提取及其模式识别应用研究

发布时间：2018-07-01 20:19

本文选题：头皮脑电 + 时频分析　；参考：《杭州电子科技大学》2012年硕士论文

【摘要】：头皮脑电(Electroencephalography, EEG)由于其无创性、易采集以及较好的时间分辨率，在神经信息工程的研究中，尤其是最近被广泛关注的脑机接口应用中，有着不可替代的作用。但EEG信号的空间分辨率较低，且具有非平稳和非线性的特性，因此寻求一种能较好刻画头皮脑电特征，并最终实现其模式识别应用的方法，受到了越来越多的关注。本文首先针对基于脑电的疲劳状态识别和预警问题，设计了促疲劳实验以及疲劳评价量表，采集了非疲劳、一般疲劳以及严重疲劳状态的头皮脑电信号，采用希尔伯特-黄变换(Hilbert-Huang Transform, HHT)时频分析方法得到其边际谱能量，给出了疲劳指数的定义和实现，最后获得了本文方法所对应的特征分布。与短时傅里叶变换(Short-time Fourier Transform，STFT)以及小波变换(Wavelet Transform, WT)相比，，HHT方法具有更好的时频分辨率；且经验模态分解(Empirical Mode Decomposition, EMD)过程，对于非平稳的脑电信号而言，更具有一种自适应能力。随后本文针对MIT-BIH生理信息库中的睡眠脑电数据，研究时频分析方法在睡眠脑电自动分期中的应用。考虑到传统HHT在进行EMD分解时，内蕴模态函数(intrinsic mode function,IMF)并不能满足单一组分模态的要求，因此本文提出将HHT时频分析方法与小波包变换(Wavelet Package Transform, WPT)相结合，应用于睡眠脑电的分期识别上。小波包变换将改善EMD分解时对信号频率“窄带”的要求，利用高低频的频带细分以及瞬时能量计算，实现对睡眠脑电的自动客观评测，分期平均正确率达到了87.37%，比未改进前的HHT方法有了较大提高。最后考虑到脑机接口技术的广泛关注，本文以国际脑机接口竞赛中的运动想象脑电模式识别为具体应用，在采用小波包和HHT时频分析方法提取事件相关去同步(Event-relatedDesynchronization, ERD)以及事件相关同步(Event-related Synchronization, ERS)特征基础上，本文提出采用一定的IMF权值优化和筛选规则，有效降低了特征提取时的高维数，以及低相关IMF对识别性能的影响。实验结果表明改进后的方法在识别ERD/ERS现象的准确性和快速性方面，有了一定的提高。
[Abstract]:Electroencephalography (EEG) plays an irreplaceable role in the research of neural information engineering, especially in the application of brain-computer interface (BCI), because of its noninvasive, easy acquisition and good time resolution. However, the spatial resolution of EEG signal is low, and it is non-stationary and nonlinear. Therefore, more and more attention has been paid to finding a method that can describe the scalp electroencephalogram and finally realize the application of EEG pattern recognition. First of all, aiming at the problem of fatigue state recognition and warning based on EEG, a fatigue promoting test and fatigue evaluation scale are designed, and scalp EEG signals of non-fatigue, general fatigue and severe fatigue state are collected. The marginal spectral energy is obtained by using Hilbert-Huang transform (HHT) time-frequency analysis method, and the definition and implementation of fatigue exponent are given. Finally, the characteristic distribution corresponding to this method is obtained. Compared with Short-time Fourier transform (STFT) and Wavelet transform (WT), HHT method has better time-frequency resolution, and empirical Mode decomposition (EMD) process is more adaptive for non-stationary EEG signals. Then, according to the sleep EEG data in MIT-BIH physiological information base, the application of time-frequency analysis method in sleep EEG automatic staging is studied. Considering that the intrinsic mode function (intrinsic mode) can not satisfy the requirement of single component mode when the traditional (intrinsic mode is decomposed, this paper proposes a new time-frequency analysis method combining with wavelet package transform (WPT). It is applied to the stage recognition of sleep EEG. Wavelet packet transform will improve the requirement of "narrow band" of signal frequency when EMD is decomposed. By using frequency subdivision of high and low frequency and calculation of instantaneous energy, automatic and objective evaluation of sleep EEG can be realized. The average accuracy of stage is 87.37, which is higher than that of HHT method. Finally, considering the extensive concern of brain-computer interface technology, this paper takes the sport imagination EEG pattern recognition in the international brain-computer interface competition as the concrete application. On the basis of extracting Event-related synchronization (ERD) and Event-related synchronization (ERS) features by wavelet packet and HHT time-frequency analysis, this paper proposes to adopt certain IMF weight optimization and screening rules to effectively reduce the high dimension of feature extraction. And the influence of low correlation IMF on recognition performance. The experimental results show that the improved method can improve the accuracy and rapidity of identifying ERD / ERS phenomenon.
【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】：2012
【分类号】：TP334.7;TP391.41

【相似文献】