基于脑电信号的情感识别

发布时间：2018-02-14 16:35

本文关键词： 脑电信号情感识别时频分析二次型时频分布 D-S证据组合理论　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：情感识别是人机交互技术中一个重要的研究领域。通过让计算机感知到人类的情感状态以及情感的变化,使其能够对人类的行为有一简单的认知识别并不断完善其人机交互的方式,提高人性化水平。在情感识别研究中,除了使用到情感语音、情感词汇及面部表情等外部行为外,人体产生的生理信号也可以用作情感识别。脑电信号(Electroencephalogram,EEG)是生理信号中一个重要的组成部分,它反映的是大脑皮层神经元的电位活动,它具有安全、实时性好等特点。由于大脑是人体的高级神经中枢,其与人体的生理活动及心理活动有着极其密切的联系。因此,可以借助于EEG来判别人的情感状态。由于大脑结构的复杂性,导致EEG具有非平稳性、随机性等特点,并且EEG还容易受到其它生理信号及外界的干扰,这些都对EEG有效成分的提取及情感的识别造成了不小困难。因此,相对其它情感识别方式而言,基于EEG的情感识别研究还尚处在起始阶段。本文首先对情感模型、EEG特征、EEG与情感的联系以及EEG分析中常用的分析方式分别作了介绍,然后针对EEG节律的提取分离以及EEG多个节律整合识别的问题,分别展开研究并取得了一定的进展。本文主要研究如下:(1)利用时频分析在非线性信号分析处理中的优势,将时频分析引入盲源分离算法中,提出基于时频盲源分离的EEG情感识别。利用时频分析能够有效地提取非平稳信号EEG中的时变局部特征以及源信号在时—频域具有的时频多样性,从而分离出观测信号中的有效成分。通过与其他方法的实验结果对比,检验了其分离的效果。(2)此外,还针对二次型时频域联合分析中常见的交叉项干扰问题,以维格纳分布为例,对交叉项抑制问题作了较为深入的分析,并运用重排理论对时频域中的能量分布作了再次分配,从而引出了维格纳分布的另一种重要改进型——重排维格纳分布,并在此基础上,引出了基于重排维格纳分布的盲源分离算法,通过实验检验了算法的分离效果。(3)针对EEG情感识别中存在的特征维度空间大、多节律整合难的问题,从信息融合的角度出发,提出将D-S证据组合理论应用于EEG的情感识别中。利用信息融合技术中经典的D-S证据组合理论,通过多个并行的BP识别网络分别作不同的单一节律情感识别,然后将每个BP识别网络作为一个独立的证据体,再通过D-S理论对各BP网络的识别结果作融合判定,最终得到决策结果。能得到比单一节律或多个节律简单整合更为准确、可靠的结果。还降低了特征空间维度,为多节律识别提供了更多的选择。此外,还针对D-S理论中出现的一些问题,对其作了改进,通过实验检验了改进后的性能。
[Abstract]:Emotion recognition is an important research field in human-computer interaction technology. So that it can recognize human behavior and improve its human-computer interaction. In the research of emotion recognition, it not only uses external behaviors such as emotional speech, emotional vocabulary and facial expression, but also improves the level of humanization. The physiological signals produced by the human body can also be used as emotional recognition. Electroencephaloencephalogram (EGG) is an important component of physiological signals, which reflects the potential activity of cortical neurons and is safe. Because the brain is the advanced nerve center of the human body, it is closely related to the physiological and psychological activities of the human body. Therefore, the EEG can be used to judge the emotional state of the human body. As a result, EEG has the characteristics of non-stationary and randomness, and EEG is vulnerable to other physiological signals and external interference, which make it difficult to extract the effective components of EEG and to recognize emotion. Compared with other affective recognition methods, the research of affective recognition based on EEG is still in its infancy. Firstly, this paper introduces the relationship between EEG features and emotion and the analysis methods commonly used in EEG analysis. Then, aiming at the problems of EEG rhythm extraction and separation and the integration of EEG rhythm recognition, the research has been carried out and some progress has been made. In this paper, we mainly study the advantages of time-frequency analysis in nonlinear signal analysis and processing, which are as follows: (1) in this paper, we make use of the advantages of time-frequency analysis in nonlinear signal analysis and processing. Time-frequency analysis is introduced into blind source separation algorithm. EEG emotion recognition based on blind time-frequency source separation is proposed. The time-frequency analysis can effectively extract the time-varying local characteristics of non-stationary signals in EEG and the time-frequency diversity of source signals in time-frequency domain. By comparing with the experimental results of other methods, the effect of separation is tested. In addition, the common crossover interference problem in quadratic time-frequency domain joint analysis is also discussed. Taking Wigner distribution as an example, the cross-term suppression problem is analyzed in depth, and the energy distribution in time-frequency domain is redistributed by using the theory of rearrangement. Thus, another important modified Wigner distribution, rearrangement Wigner distribution, is introduced. On this basis, a blind source separation algorithm based on rearrangement Wigner distribution is proposed. The separation effect of the algorithm is tested by experiments. (3) aiming at the problems of large space of feature dimension and difficult integration of multi-rhythm in EEG emotion recognition, this paper starts from the angle of information fusion. The D-S evidence combination theory is applied to the emotion recognition of EEG. Using the classical D-S evidence combination theory in information fusion technology, different single rhythm emotion recognition is made through multiple parallel BP recognition networks. Then each BP recognition network is regarded as an independent evidence body, and then the recognition results of each BP network are fused and judged by D-S theory. Finally, the decision result is obtained, which is more accurate than the simple integration of single rhythm or multiple rhythms. Reliable results. It also reduces the dimension of feature space and provides more choices for multi-rhythm recognition. In addition, some problems in D-S theory are improved, and the improved performance is tested by experiments.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【相似文献】