计算机工程 ›› 2019, Vol. 45 ›› Issue (4): 196-204.doi: 10.19678/j.issn.1000-3428.0050486

• 人工智能及识别技术 • 上一篇    下一篇

基于脑电信号瞬时能量的情感识别方法

陈田1a,1b,陈占刚1a,1b,袁晓辉2,鞠思航1a,1b,任福继1a,1b,3   

  1. 1.合肥工业大学 a.计算机与信息学院; b.情感计算与先进智能机器安徽省重点实验室,合肥 230009; 2.北德克萨斯州大学 计算机与工程学院,美国 丹顿市 76203; 3.德岛大学 工学部,日本 德岛 770-8506
  • 收稿日期:2018-02-11 出版日期:2019-04-15 发布日期:2019-04-15
  • 作者简介:陈田(1974—),女,副教授,主研方向为情感计算;陈占刚,硕士研究生;袁晓辉,副教授;鞠思航,硕士研究生;任福继,教授。
  • 基金项目:

    国家自然科学基金重点项目(61432004);国家自然科学基金(61204046,61474035);国家留学基金(201706695016)。

Emotion Recognition Method Based on Instantaneous Energy of Electroencephalography

CHEN Tian1a,1b,CHEN Zhan’gang1a,1b ,YUAN Xiaohui2,JU Sihang1a,1b,REN Fuji1a,1b,3   

  1. 1a.School of Computer and Information;1b.Anhui Province Key Laboratory of Affective Computing and Advanced Intelligent Machine,Hefei University of Technology,Hefei 230009,China; 2.School of Computer and Engineering,University of North Texas,Denton 76203,USA; 3.Faculty of Engineering,The University of Tokushima,Tokushima 770-8506,Japan
  • Received:2018-02-11 Online:2019-04-15 Published:2019-04-15

摘要:

希尔伯特-黄变换(HHT)是一种处理脑电信号(EEG)的有效方法,包括经验模态分解(EMD)和Hilbert变换2个部分。但EMD无法分解包含低能量的信号,且在低频区域会产生不良的本征模态函数。为消除EMD的弊端,提出一种小波包变换(WPT)和HHT相结合的EEG处理方法。采用WPT将EEG分解成一组窄带信号,通过HHT得到Hilbert能量谱,求出平均瞬时能量作为EEG特征并封装成特征矩阵。将特征矩阵通过卷积神经网络(CNN)、递归神经网络(RNN)、支持向量机(SVM)组成的混合情感识别模型进行训练与分类。实验结果表明,该方法对高兴、悲伤、平静、恐惧4种情感的平均识别率为86.22%,最优识别率为93.45%。

关键词: 脑电信号, 情感识别, 希尔伯特-黄变换, 卷积神经网络, 递归神经网络

Abstract:

Hilbert-Huang Transform(HHT) is an effective method to deal with Electroencephalography(EEG) that includes two parts:Empirical Mode Decomposition(EMD) and Hilbert transform.However,the EMD cannot decompose a signal of low energy and will produce bad Intrinsic Mode Functions(IMF) in the low frequency region.To eliminate the effects of EMD,this paper proposes an EEG processing method which combines Wavelet Packet Transform(WPT) and HHT.Firstly,the EEG is decomposed into a set of narrow-band signals by WPT,the Hilbert energy spectrum of EEG is obtained by HHT,and the average value of the instantaneous energies is calculated as the EEG feature and packaged into a feature matrix.The feature matrix is trained and classified by mixed emotion recognition model which consists of Convolutional Neural Network(CNN),Recurrent Neural Network(RNN),and Support Vector Machine(SVM).Experimental results show that the average recognition rate and the best recognition rate of the four emotions which are happiness,sadness,calmness,and fear are 86.22% and 93.45%.

Key words: Electroencephalography(EEG), emotion recognition, Hilbert-Huang Transform(HHT), Convolutional Neural Network(CNN), Recurrent Neural Network(RNN)

中图分类号: