基于EEG信号后段正电位特征的情绪识别

发布时间：2018-01-05 17:41

本文关键词：基于EEG信号后段正电位特征的情绪识别　出处：《郑州大学》2017年硕士论文　论文类型：学位论文

【摘要】：如何准确的解读人类的情绪状态一直是人工智能、医学、计算机等领域的科研工作者急需解决的问题,在情绪识别研究领域中,研究学者不断努力已经取得了一些进展,但是,还需要我们不断努力。目前,基于脑电信号的情绪识别研究热度与日俱增。非植入式电极的应用让脑电信号的采集更加方便与快捷,也让基于脑电信号的相关成果应用到生活中成为可能。由于目前人们的工作与生活的压力比较大,存在很多潜在的心理上的问题,日久天长,为身体健康埋下隐患。基于脑电信号的情绪识别研究能够识别个体脑电信号中的情绪种类,即使表面上个体的情绪被伪装,但是脑电信号能够真实反映个体的精神状态,为接下来的调整提供帮助,解决潜在的问题,为自己的身体健康与精神健康提供保障。同时,识别人脑电信号中的情绪也人机交互的情感智能和疾病的治疗提供了一种方法。脑电信号的情绪识别研究已有一段时间,主要研究的方法集中在在时域、频域、时频、统计学等特征分析和离散、连续等分类模型等方面,鲜有人关注情绪发生的脑电本质特征。脑电信号中的后段正电位(300~1500ms)发生在刺激呈现后的几百毫秒,持续几秒钟,ERP后段正电位参与情绪的加工过程,其对情绪的影响作用最大的脑区位置是在在中央-顶区,ERP中的后段正电位是重要的情绪识别特征。本文将研究人们忽视的脑电信号中的晚期成分,探究脑电信号后段正电位特征识别脑电信号中的情绪的效果。本文主要的工作如下:首先,采集脑电信号,并对脑电信号做特征识别与分析前的预处理。在这一部分主要包括脑电信号采集实验的设计、被试者的选择与情绪诱发实验的进行并记录脑电信号。本文的实验根据研究内容设计以图片材料诱发被试产生正性、中性、负性三种情绪。采集到脑电信号后,对脑电信号进行了预处理,主要运用独立分量分析方法去除脑电信号中的各种伪迹与噪声,并最大程度保留脑电成分。其次,提取脑电信号特征并分析。本文提取了脑电信号中的后段正电位特征,并提取了三个时频特征进行对比分析。通过对平均事件相关电位的方差分析发现,在某些导联上,后段正电位早期(300 ms~600 ms)、中期(600ms~1000 ms)和晚期(1000 ms~1500 ms)具有显著性差异,被试者的后段正电位的早期(300 ms~600 ms)在枕叶区、颞叶区以及顶叶区的三种情绪的情绪特征差异明显,说明被试者的后段正电位参与了被试者的情绪反应和调节,后段正电位特征能够对正性、中性、负性情绪进行分类。最后,对脑电信号中的情绪进行分类识别并分析。利用支持向量机算法和K近邻分类算法对两类情绪特征进行分类测试。后段正电位的早期(300~600ms)、中期(600~1000ms)以及晚期(1000~1500ms)特征能够识别脑电信号中的正性、中性以及负性情绪种类,其中β频段内的后段正电位的晚期特征识别准确率最高,达到83.33%,说明脑电信号的后段正电位能够作为基于脑电信号的情绪种类识别的特征。为了进一步说明问题,本文还提取了三种时频特征进行脑电信号的情绪种类识别做为对比来分析,脑电信号后段正电位特征的情绪识别准确率比时频特征的情绪识别准确率高。可以证明本文探究的脑电信号的后段正电位特能够识别脑电信号中的正性、中性、负性情绪种类,为脑电信号的情绪种类识别提供了一种方法。
[Abstract]:How to accurately interpret the emotional state of humans is artificial intelligence, medical, computer and other fields of scientific research workers need to solve the problem, in the emotion recognition research field, scholars continue to have made some progress, but we still need to continue efforts. At present, the heat of emotion recognition of EEG signals based on the grow with each passing day application of non implantable electrodes. So EEG signal acquisition faster and more convenient, but also to the relevant results based on EEG signal to life possible. Due to the people's life and work pressure is relatively large, there are many potential psychological problems, for many, many years to come for the health danger,. Emotion recognition of EEG signals to identify individual EEG emotion types based on, even on the surface of individual mood be disguised, but EEG signal can be true Reflect the individual's mental state, to provide help for the next adjustment, to solve the potential problems for their physical health and mental health security. At the same time, provides a method for identification of electroencephalogram in emotion interactive emotional intelligence and disease. The EEG has a study of emotion recognition time, the main research method in time domain, frequency domain, time-frequency characteristics, statistical analysis and discrete, continuous classification model, people pay little attention to the essential characteristics of EEG emotion. After EEG positive potentials (300~1500ms) in the hundreds of milliseconds after stimulus, lasting a few second, the process of ERP after the positive potential brain areas involved in emotional, position of the largest influence on the role of emotions in the central parietal region, after ERP in the positive potential is an important emotion recognition feature. In this paper The advanced components of EEG research neglected in the study of EEG after normal EEG feature recognition potential in the emotion effect. The main work of this paper is as follows: firstly, the EEG signal acquisition, preprocessing and recognition and analysis of the characteristics of the EEG in the design of this part. Including the EEG signal acquisition experiment, subjects selection and emotional experiments conducted and recorded EEG signals. The experimental design according to the research content positively, to picture materials induced by three kinds of neutral subjects, negative mood. The collected EEG, EEG. The pretreatment, mainly using independent component analysis method to remove various artifacts and noise in EEG, and retain the greatest degree of brain electrical components. Secondly, EEG feature extraction and analysis. In this paper, the extraction of EEG in posterior segment positive characteristics A comparative analysis, and extract the frequency characteristics of three. The average variance of event-related potential analysis found that in some leads, after the early period of positive potential (300 ms~600 MS), mid (600ms~1000 MS) and late (1000 ms~1500 MS) with significant differences, the early period of subjects positive potential (300 ms~600 MS) in the occipital region, differences in emotional characteristics of three kinds of emotions and temporal parietal regions is obvious, that the subjects of the posterior segment is involved in emotional responses to potential participants and adjust, after the positive potential characteristics can be of positive, neutral, negative emotion classification. Finally, classification and analysis of EEG in the mood. The support vector machine algorithm and K neighbor classification algorithm to classify the test for two types of emotional features. Early after a positive potential (300~600ms), middle (600~1000ms) and late (1000~1500ms) feature can Positive identification of EEG, neutral and negative emotion types, including advanced feature recognition after the beta band positive potential, the highest accuracy rate reached 83.33%, the segment of EEG positive potentials can be used as features of emotion identification based on EEG. In order to further illustrate the problem, this paper also extracts the frequency characteristics of the emotion recognition of EEG signals of three kinds of the contrast analysis, emotion recognition of emotion recognition of EEG after positive potential characteristics of accuracy of characteristic frequency ratio when high accuracy. After EEG signals can prove that this paper explores the positive potential to positive characteristics identification, EEG neutral, negative emotion categories, provides a method for recognition of emotional EEG signals.

【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

【相似文献】