Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (4): 191-196.DOI: 10.3778/j.issn.1002-8331.2109-0083

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Research on Emotion Recognition Based on EEG Time-Frequency-Spatial Multi-Domain Feature Fusion

WANG Lu, LIANG Mingjing, SHI Huiyu, WEN Xin, CAO Rui   

  1. Department of Software, Taiyuan University of Technology, Taiyuan 030024, China
  • Online:2023-02-15 Published:2023-02-15



  1. 太原理工大学 软件学院,太原 030024

Abstract: The traditional emotion recognition based on electroencephalogram(EEG) mainly adopted a single EEG feature extraction approach. In order to make full use of the rich information contained in EEG, a new method of EEG emotion recognition based on multi-domain feature fusion is proposed. This paper extracts EEG features in time-domain, frequency-domain and space-domain, and fuses the three domain features as the input of the emotion recognition model. Firstly, the power spectral density of the three frequency bands of alpha, beta and gamma of the EEG signal in different time windows are calculated, and combined with the spatial information of the EEG electrode, the EEG images are formed. Then, the convolutional neural network(CNN) and bidirectional long short-term memory network(BLSTM) are used to construct the CNN-BLSTM model for emotion recognition, and the features of time, frequency and space domains are learned respectively. The method is verified in the SEED dataset. The results show that the method can effectively improve the accuracy of recognition, and the average recognition accuracy is 96.25%.

Key words: electroencephalogram(EEG), feature extraction, CNN-BLSTM, emotion recognition

摘要: 传统基于脑电信号(electroencephalogram,EEG)的情感识别主要采用单一的脑电特征提取方法,为了充分利用EEG中蕴含的丰富信息,提出一种多域特征融合的脑电情感识别新方法。提取了EEG的时域、频域和空域特征,将三域特征进行融合作为情感识别模型的输入。首先计算不同时间窗EEG信号的alpha、beta、gamma三个频段功率谱密度,并结合脑电电极空间信息构成EEG图片,然后利用卷积神经网络(convolutional neural network,CNN)与双向长短期记忆网络(bidirectional long short-term memory network,BLSTM)构建CNN-BLSTM情感识别模型,分别对时、频、空三域特征进行学习。在SEED数据集对该方法进行验证,结果表明该方法能有效提高情感识别精度,平均识别准确率达96.25%。

关键词: 脑电信号(EEG), 特征融合, CNN-BLSTM, 情感识别