【作者】 徐金燕；

【导师】 吴小培；

【作者基本信息】 安徽大学 ， 信号与信息处理， 2005， 硕士

【摘要】 事件相关电位(Event-Related Potential,ERP)反映了认知过程中大脑的神经电活动,是研究人类高级认知功能的一个有力工具,在临床医学及认知科学中获得了广泛的应用。然而ERP信号非常微弱,常常淹没在自发脑电之中,此外还极易受到其它各种噪声的干扰,如眼电(EOG)、肌电(EMG)、心电(ECG)以及各种环境噪声等,因此长期以来ERP信号处理的重点之一就集中在如何从强背景噪声下提取有用的ERP信号。 独立分量分析(Independent Component Analysis,ICA)是近年发展起来的一种新的多维信号处理方法,它的研究对象是相互独立的非高斯信号。在满足一定的条件下,ICA能够从同步测量的多道观测信号中分离出隐含的独立信源。 本文研究了ICA的基本理论及其算法,并将其应用于ERP的消噪及P3亚成分的提取,且在以下几个方面做了一些有特色的工作: 1)对独立分量分析算法进行了研究,重点讨论了基于非高斯性极大原理和信息极大原理的两类有代表性的ICA算法。 2)对独立分量分析在ERP消噪中的应用进行了研究。首先对实测的视觉诱发ERP数据运用ICA算法进行分解,然后利用脑电的先验知识,从时域、频域及空间分布模式入手对ICA分离出的独立分量进行噪声识别。这种多角度分析法避免了仅从某一个角度分析时所带有的经验性。此外,传统的基于ICA的ERP消噪通常只对非神经电活动(如眼电、肌电等)噪声进行消除,本文除进行了这方面的工作之外,还对部分与刺激具有锁时(time-lock)关系的自发脑电噪声(如α波、μ波)也进行了消除,从而克服了在后续的ERP提取过程中所造成的难以抑制这些锁时噪声的缺陷。实验获得了理想的ERP消噪效果,并为ERP的可靠提取提供了一条切实可行的新思路。 3)对ICA在P3亚成分提取中的应用进行了研究。利用ICA可将混合在观测信号中的相互独立的源信号分离出来的特性,成功地提取出了P3复合波中的各亚成分。这些亚成分的提取可以帮助我们对人类认知等高级神经活动进行更深入的探索。更多还原

【Abstract】 Event-Related Potential(ERP) reflects neural activity of the brain of cognitive process, and is a useful tool of studying human advanced cognitive function, so it has a widely application in clinical medicine and cognitive science. But ERP signal is often heavily contaminated by sophisticated background noise such as spontaneous Electroencephalogram(EEG), Electrooculogram(EOG), Electromyogram(EMG), Electrocardiogram(ECG) etc., and is very weak compared with the background noise in which it is embedded. For this reason, it is necessary to extract ERP signal from the strong background noise.Independent Component Analysis(ICA) is a novel multi-dimensional signal processing method developed recently and used to analyze the mutually independent nongaussian signals. When some certain assumptions are satisfied, ICA can effectively separate the underlying independent sources only from the synchronous multichannel observed mixtures.This thesis studies the theories and algorithms of ICA and explores its applications in ERP denoising and P3 subcomponents extraction. The innovated works we have finished are as follows:1) Studying the two typical algorithms of ICA based on the nongaussian maximum principle and information maximum principle .2) Studying the ICA-based method to remove EEG artifacts in order to efficiently extract ERP signal from the strong background noise. In the experiments we analyze the EEG independent components separated by ICA algorithm, by means of the time analysis, frequency analysis and scalp topography analysis, to find the noise components. The manifold analytical method enhances the reliability of the components analysis. Moreover, the traditional ERP denoising by ICA is commonly confined to remove non-neural artifacts such as EOG and EMG, but in this thesis, we remove not only EOG and EMG but also the time-lock spontaneous EEG such as a rhythms and μ. rhythms in order to obtain better ERP denoising effect. The experimental results show, by the new ERP denoising method, ERP signal is efficiently extracted from the strong background noise.3) Studying P3 subcomponents extraction based on ICA. According to the properties that ICA can separate the underlying independent sources from the observed mixtures, P3 subcomponents are successfully extracted from the P3 complex. Study of those subcomponents can help people to further explore human advanced cognitive function, and make the study of P3 come up to higher level.更多还原