计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (22): 99-105.DOI: 10.3778/j.issn.1002-8331.1807-0219

• 模式识别与人工智能 • 上一篇    下一篇

基于改进的卷积神经网络脑电信号情感识别

田莉莉,邹俊忠,张见,卫作臣,汪春梅   

  1. 1.华东理工大学 信息科学与工程学院 自动化系,上海 200237
    2.上海师范大学 信息与机电工程学院 自动化系,上海 200234
  • 出版日期:2019-11-15 发布日期:2019-11-13

Emotion Recognition of EEG Signal Based on Improved Convolutional Neural Network

TIAN Lili, ZOU Junzhong, ZHANG Jian, WEI Zuochen, WANG Chunmei   

  1. 1.Department of Automation, School of Information Science and Engineering, East China University of Science and Technology, Shanghai 200237, China
    2.Department of Automation, School of Information Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 200234, China
  • Online:2019-11-15 Published:2019-11-13

摘要: 针对传统机器学习需要人工构建特征及特征质量较低等问题,提出一种新颖的基于一维卷积神经网络(Convolutional Neural Network,CNN)的特征提取方法。采用编码思想,由卷积层和下采样层构成编码器网络提取脑电信号情感特征,随后与特征图一起输入Leaky ReLU激活函数。对于卷积预训练过程,使用交叉熵和正则化项双目标优化损失函数,之后采用随机森林分类器以获得情感分类标签。在国际公开数据集SEED上进行实验,达到94.7%的情感分类准确率,实验结果表明了该方法的有效性和鲁棒性。

关键词: 脑电信号(EEG), 特征提取, 卷积神经网络(CNN), 随机森林, 损失函数

Abstract: Considering that traditional machine learning requires artificial construction features and low feature quality, this paper proposes a novel automatic feature extraction approach in Electroencephalograph(EEG) signals based on 1-D Convolutional Neural Network(CNN). This approach uses the idea of compilation, at the same time the convolutional layer and the downsampling layer form the encoder network to extract the emotional characteristics of the EEG signal, then the Leaky ReLU activation function is applied to the feature map. For the convolution pre-training process, the cross-entropy and regularization terms are used to optimize the loss function, then the random forest classifier is used to obtain the emotion classification label. Finally, the experiment is carried out on the international public data set SEED, which achieves 94.7% sentiment classification accuracy, and the experimental results show the effectiveness and robustness of the proposed method.

Key words: Electroencephalograph(EEG), feature extracting, Convolutional Neural Network(CNN), fandom forest, loss function