基于脑电信号的情感识别研究

发布时间：2018-06-05 05:55

本文选题：情感识别 + 分形维数　；参考：《南京邮电大学》2017年硕士论文

【摘要】：如今有关脑电信号(EEG)的相关研究已经成为脑机接口(BCI)、人工智能(AI)、计算机科学及医疗卫生等领域的热门话题,其中基于脑电信号的情感识别研究正日趋火热,无论是将情感识别应用人工智能,还是用于辅助抑郁症及自闭症等精神类疾病发病机制的研究和治疗上,都具有重要的价值和意义。本文将脑电信号的分形特性引入情感识别的研究中,即将脑电信号的Higuchi维数作为特征值以区分不同的情感状态。具体的研究内容划分为三大部分,第一部分,设计实验,收集实验素材,诱发实验对象让其处于不同的情感状态,并获取对应的EEG数据;第二部分,计算脑电信号的Higuchi维数(HF)值,并以此作为特征值分别算出两类支持向量机(SVM)分类算法下的分类准确率;第三部分,基于第二部分的分析结果,进一步运用脑电信号的Higuchi维数特征,并配合三种分类算法,分析与情感状态关联强的频率段和脑部区域。第二部分在获取分类准确率之前,选用主成分分析(PCA)对62导联EEG信号实现降维,并最终保留了6导联脑电信号,每一导联信号经运算共计得到10组HF值,各取愉悦、平静、悲伤状态下的一个HF值构成三维的特征向量,通过两类支持向量机和10层交叉验证法得出最后的分类准确率,最终结果是线性核支持向量机获得最高83.33%的分类准确率,至此,表明EEG信号的分形特性可用于区分脑电信号所代表的情感状态的差异。第三部分,考虑到脑电信号情感识别将来会运用到实际的生活中,而使用全部的64个数据采样点进行数据采样不太态现实,因此减少采样点的数目势在必行,同时鉴于脑电信号固有的节律特性,本章继续借助HF值来寻找与情感关联度最大的脑部区域和频率段,选用支持向量机、素贝叶斯以及K近邻分析作为分类算法,最终得到12位实验对象的累计准确率,数据分析结果表明,频率较高的Alpha、Beta以及Gamma频段获得累计准确率明显高于频率较低的频率段,其次,在单独分析累计准确率相对稳定的Beta频段,试图找寻特定脑部区域时,分析发现额区、颞区及其周边采样点的HF值具有较高的累计分类准确率,至此可证明EEG信号的分形特性可用于情感状态的识别。
[Abstract]:Nowadays, the research on EEG has become a hot topic in the fields of brain-computer interface (BCI), artificial intelligence (AI), computer science, medical and health, etc. Among them, the research on emotion recognition based on EEG is getting hotter and hotter. It is of great value and significance not only to apply artificial intelligence to emotion recognition, but also to study and treat the pathogenesis of mental diseases such as depression and autism. In this paper, the fractal characteristics of EEG signals are introduced into the study of emotion recognition. The Higuchi dimension of EEG signals is taken as the eigenvalue to distinguish different emotional states. The specific research contents are divided into three parts. The first part is to design experiments, collect experimental materials, induce experimental objects to be in different emotional states, and obtain corresponding EEG data. The Higuchi dimension of EEG is calculated and the classification accuracy of two kinds of support vector machine (SVM) classification algorithms is calculated as the eigenvalue. The third part is based on the analysis results of the second part. Using the Higuchi dimension feature of EEG signal and three classification algorithms, the frequency segment and brain region which are strongly correlated with emotional state are analyzed. In the second part, before the classification accuracy is obtained, principal component analysis (PCA) is used to reduce the dimension of 62-lead EEG signal, and finally the 6-lead EEG signal is retained. After operation, 10 groups of HF values are obtained for each lead signal, each of which is pleasurable and calm. A HF value in a sad state constitutes a three-dimensional feature vector. The final classification accuracy is obtained by two kinds of support vector machines and a 10-layer cross-validation method. The final result is that the linear kernel support vector machine obtains the highest classification accuracy of 83.33%. It shows that the fractal characteristics of EEG signal can be used to distinguish the difference of emotional state represented by EEG signal. In the third part, considering that EEG emotion recognition will be used in real life in the future, it is not very realistic to use all 64 data sampling points to sample data, so it is imperative to reduce the number of sampling points. At the same time, in view of the inherent rhythmic characteristics of EEG signals, this chapter continues to use HF value to find the regions and frequencies of the brain with the greatest affective correlation, and selects support vector machines, Bayes and K-nearest neighbor analysis as classification algorithms. Finally, the cumulative accuracy of 12 experimental subjects was obtained. The results of data analysis showed that the accumulative accuracy of Alpha-Beta and Gamma bands with higher frequency was significantly higher than that of the frequency band with lower frequency. When the Beta frequency band with relatively stable accumulative accuracy was analyzed separately, the HF values of frontal region, temporal region and their peripheral sampling points were found to have higher accumulative classification accuracy when trying to find specific brain regions. It is proved that the fractal characteristics of EEG signal can be used to identify the emotional state.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TP18

【相似文献】