情绪图片视觉诱发EEG特征提取与分析

发布时间：2018-07-13 15:19

【摘要】：1872年,达尔文在《人类和动物的表情》一书中指出情绪是高级进化阶段的适应工具,从此人们开始了情绪实验与理论的研究。经过100多年,到20世纪后期情绪研究蓬勃发展起来,并与认知、神经科学、脑科学等研究相结合;其研究手段也多种多样,如脑电(EEG)、功能磁共振成像(fMRI)、功能近红外成像(fNIRI)等。EEG因其高时间分辨率和简便易行优势,被广泛用于情绪研究中。本文设计了基于国际标准情绪图片库(IAPS)的情绪图片视觉诱发实验,被试者观看各等级的情绪图片并采集EEG信号。通过对EEG信号进行特征提取与分析,找到与情绪变化相关的EEG特征,并尝试在EEG特征与情绪等级之间建立对应关系,以期实现基于脑电特征的情绪等级分类识别。文中首先对被试者观看图片时的EEG进行功率谱分析,构建其功率谱脑地形图。由该地形图可知,情绪图片视觉诱发时前额区域脑电最为活跃。信号的频谱分析表明EEG能量主要集中在15Hz以下。为找到EEG信号最具情绪可分性的频段,本文对一些导联的EEG进行了可分频段分析;同时,对EEG信号进行了功率谱熵、相关维数分析,并对AF3、AF4、F3、F4导联的EEG特征进行了最小二乘直线拟合,在情绪等级与EEG特征之间建立了对应关系。在模式识别环节,首先分别使用支持向量机的5-折交叉验证方法和隐马尔科夫模型对所提取的脑电信号特征进行了分类识别;然后进行了特征层融合后的模式识别,得到融合特征的分类识别率。结果显示,特征信息融合后,本文对情绪图片等级一、五、八的最高平均识别率达到86.5%。目前已经能够通过情绪图片诱发EEG更客观的将最消极、中性、最积极这三种情绪状态区分开,下一步将进一步研究将每个等级区分开的特征提取与分类识别算法。
[Abstract]:In 1872, Darwin pointed out in Human and Animal expressions that emotion is an adaptive tool in the advanced stage of evolution, from which people began to study emotional experiments and theories. After more than 100 years, by the late 20th century, emotional research has flourished, and combined with cognitive, neuroscience, brain science, and so on. EEG, such as EEG, functional magnetic resonance imaging (fMRI), functional near infrared imaging (fNIRI) and so on, are widely used in emotional research because of their high temporal resolution and simplicity. Based on the International Standard emotional Picture Library (IAPS), a visual evoked experiment of emotion picture was designed in this paper. The subjects watched the emotion pictures of different levels and collected EEG signals. Through the feature extraction and analysis of EEG signals, the EEG features related to emotional changes are found, and the corresponding relationship between EEG features and emotion grades is attempted to be established, in order to realize the classification and recognition of emotion grades based on EEG features. In this paper, the EEG of the subjects watching the picture is analyzed by power spectrum analysis, and the brain map of the power spectrum is constructed. According to the topographic map, the frontal area is the most active when the emotional picture is visually induced. Spectrum analysis shows that EEG energy is mainly below 15 Hz. In order to find out the most emotional band of EEG signal, this paper analyzes the frequency band of EEG with some leads, and analyzes the power spectrum entropy and correlation dimension of EEG signal. The EEG features of AF3 / AF4 / F3F4 lead were fitted by least-square straight line fitting, and the corresponding relationship between emotional grade and EEG features was established. In pattern recognition, support vector machine (SVM) 5- fold cross validation method and hidden Markov model are used to classify the extracted EEG features, and then the feature layer fusion pattern recognition is carried out. The classification recognition rate of fusion features is obtained. The results show that after feature information fusion, the highest average recognition rate of emotional image grades 1, 5 and 8 is 86.5%. At present, EEG can be induced by emotional images to more objectively distinguish the three most negative, neutral and active emotional states. The next step will be to further study the feature extraction and classification recognition algorithm.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R318.0

【参考文献】