作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (12): 95-102. doi: 10.19678/j.issn.1000-3428.0060195

• 人工智能与模式识别 • 上一篇    下一篇

基于位置信息重建与时频域信息融合的脑电信号情感识别

柳素红, 孙晓, 李春彬   

  1. 合肥工业大学 计算机与信息学院, 合肥 230601
  • 收稿日期:2020-12-01 修回日期:2021-01-05 发布日期:2021-01-25
  • 作者简介:柳素红(1994-),女,硕士研究生,主研方向为情感计算;孙晓,教授;李春彬,硕士研究生。
  • 基金资助:
    国家自然科学基金(61976078)。

Emotion Recognition Using EEG Signals Based on Location Information Reconstruction and Time-Frequency Information Fusion

LIU Suhong, SUN Xiao, LI Chunbin   

  1. School of Computer Science and Information Engineering, Hefei University of Technology, Hefei 230601, China
  • Received:2020-12-01 Revised:2021-01-05 Published:2021-01-25

摘要: 脑电信号由中枢神经系统产生,具有很高的真实性,但存在数据量少和数据复杂等问题。为提高脑电信号情感识别准确率,在脑电信号功率谱密度的基础上提出一种脑电位置信息重建的方法,使神经网络模型可以直接获取脑电信号中不易学习的位置信息。运用融合网络从原始的脑电信号中分别抽取时域特征和频域特征,根据频域信息重建脑电信号的位置信息,将时频域信息及位置信息进行融合,以获得更高的脑电信号情感分类准确率。在公开数据集DEAP上的实验结果表明,Valence和Arousal的二分类准确率分别达到86.31%和85.57%,与传统脑电信号情感识别方法相比,该方法分类准确率得到有效提高。

关键词: 情感识别, 脑电信号, 功率谱密度, 位置信息重建, 融合网络

Abstract: Electroencephalogram(EEG) is produced by the central nervous system, and can reflect the emotions of human beings.However, the existing EEG data is insufficient and of high complexity.To improve the accuracy of emotion recognition using EEG, a method for reconstructing EEG location information based on the Power Spectral Density(PSD) of EEG is proposed, which enables the neural network model to directly obtain the EEG location information that is hard to learn.In addition, a fusion network is used to extract the time-domain features and frequency-domain features from original EEG signals, so the location information of EEG is reconstructed according to the frequency-domain information.Finally, the time-frequency domain information and location information are fused to improve the classification accuracy of EEG-based emotion recognition.The method achieves a binary classification accuracy of 86.31% for Valence and 85.57% for Arousal on an open data set.Compared with previous studies on EEG-based emotion recognition, the proposed method significantly increases the classification accuracy.

Key words: emotion recognition, Electroencephalogram(EEG) signals, Power Spectral Density(PSD), location information reconstruction, fusion network

中图分类号: