【作者】 王艳超；

【导师】 杨立才；

【作者基本信息】 山东大学 ， 生物医学工程， 2013， 硕士

【摘要】 作为涉及多领域知识的跨学科的综合类课题,情感研究目前已成为心理学、计算机科学以及医学界广泛关注的热点,它具有高度的综合性以及实际应用价值。基于生理信号的情感研究是当前心理学及医学研究领域较为新颖的课题。脉搏作为人体一种常见的生理特征,其在心脏及动脉血管系统的相互影响下能够反映动脉血压的变化情况,并且与人体的健康状况密切相关。通过提取脉搏信号不同特征来分析和辨别不同情感状态的研究可以为临床医学、社会科学以及工程应用提供理论依据与实践基础。本研究在现有生理特征情感识别研究的基础上,针对现有生理信号情感识别方法中存在的不足,提出了一种新的基于脉搏信号的情感识别方法。该方法通过提取脉搏信号的频谱特征、非线性特征和经验模态分解特征来辨别个体的不同情感状态。实验结果表明该方法对于所研究的四种基本情感状态具有较高的识别率。具体内容如下：(1)根据脉搏信号产生的生理学机理,详细分析了脉搏信号情感识别研究的可行性。脉搏作为人体的一种生理特征,其变化不受人的主观意识控制,能够客观地反映个体的情感状态。另外,脉搏信号具有采集过程简单、信号易处理等优势。(2)分析小波变换理论,结合脉搏信号在采集过程中存在的噪声类型,选用db5小波对脉搏信号进行去噪处理。对于脉搏信号的低频噪声,采用近似分量置零操作；对于脉搏信号高频噪声,采用改进的阈值处理方法,然后用处理后的小波分解系数重构脉搏信号。(3)将经验模态分解算法引入到脉搏信号情感识别研究中,针对该算法在信号分解过程中存在的端点效应提出了种改善方法。该方法首先利用信号相干平均技术获取脉搏信号模板,然后将信号端点处的数据与模板匹配,将原始信号延拓至整周期,最后利用镜像延拓法对延拓后的信号进行经验模态分解。(4)本研究将通过经验模态分解得到的模态能量商以及脉搏信号非线性特征和频谱特征融合应用于脉搏情感识别研究中,探讨了不同情感状态下脉搏信号对应的特征信息。给出了三种特征信息的提取方法,并实现脉搏信号有效特征的提取。(5)为完成不同情感状态的自动判别,本研究构造了一种基于支持向量机多分类方法,对80个样本(快乐状态20例,愉悦状态20例,悲伤状态20例,愤怒状态20例)的脉搏信号进行情感状态自动判别分析,实验结果表明本文选取的脉搏信号多模态特征信息组合对于情感状态识别效果更佳更多还原

【Abstract】 Emotion research as a comprehensive topic involving many different fields of knowledge has attracted widely attention in Psychology, Computer science and Medical field. It has the highly integration and practical value. Emotion research based on physiological signals is a novel topic in the field of Psychology and Medical. Pulse as a common physiological feature reflects the arterial pressure changes under the interaction of heart and arterial vascular system and also closed related to the health status. The difference character of pulse signals can be used to analyze and recognize vary emotion status, it is as a research can provide the basis of clinical medicine, social science and engineering application.This research is based on previous study, it puts forward a new emotion recognition method due to the deficiencies existing current emotion recognition method. It can recognize the different emotion states of individual by the extracting the spectrum character, nonlinear character and empirical mode decomposition character. It proves to be that this method has the higher recognition rate. More details as below:(1) It analyzes the feasibility of pulse signal-emotion recognition research in details according to the physiological mechanism generated by pulse signal. Pulse as a physiological feature can not be controlled subjectively and can reflect emotion objectively. In addition, pulse signals have the advantages of simple collecting and easy processing.(2) Analyzing wavelet conversion theory:remove the noise exist in the pulse signals by db5wavelet combing with the noise type in the acquisition process of pulse signals. Low-frequency noise from pulse signal processes by approximate component zero setting and high-frequency noise from pulse signal by improved threshold value and then rebuilds pulse signals by wavelet coefficients.(3) Empirical mode decomposition algorithm is introduced to the pulse signal emotion recognition study; it comes up with an improved method in the terms of endpoint effect in the process of signal decomposition. This method first takes use of signal coherent averaging technique to obtain the pulse signal template and then match the data located the endpoint with template extending original signal to the whole cycle; at last compose the extended signal in the empirical mode by Mirror extension method.(4) This study integrates EMD generated by Empirical mode decomposition and nonlinear and spectrum character of pulse signal and then apply to the pulse-emotion recognition research discussing different pulse signal reflected by different emotions。 Gained the extraction methods of three types of certain character information and released the extraction of pulse signal effective character.(5) This study also constructed Multi-classification based on supporting Vector Machine in order to fulfill auto-recognition of different emotion. Extract pulse signal from80samples(20samples of states of happiness,20samples of states of pleasure,20samples of states of sadness,20samples of states of anger) to do auto-recognition analysis and it proves that combination of pulse signal multi-mode character information is a more efficient way to emotion recognition.更多还原