计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (24): 110-116.DOI: 10.3778/j.issn.1002-8331.1808-0404

• 模式识别与人工智能 • 上一篇    下一篇

基于CBLSTM算法的脑电信号特征分类

胡章芳,崔婷婷,罗元,张毅,魏博   

  1. 1.重庆邮电大学 光电工程学院,重庆 400065
    2.重庆邮电大学 先进制造学院,重庆 400065
  • 出版日期:2019-12-15 发布日期:2019-12-11

EEG Signal Classification Algorithm Based on CBLSTM

HU Zhangfang, CUI Tingting, LUO Yuan, ZHANG Yi, WEI Bo   

  1. 1.School of Optoelectronic Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
    2.School of Advanced Manufacturing Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
  • Online:2019-12-15 Published:2019-12-11

摘要: 由于传统的脑电信号分类方法识别率较低,且识别率随着脑电信号类别的增加逐渐下降,针对脑电信号时空特征结合的特点,设计了一个多层的卷积双向LSTM型递归神经网络(CBLSTM)分类模型。此分类模型利用多层的卷积神经网络有效提取脑电序列的频域特征,采用双向LSTM提取脑电信号的时域特征,并将脑电信号序列逐帧输入到此分类模型中进行标记,最后输出分类结果。对比研究验证了所提出方法的可行性,实验表明此分类模型平均分类识别率得到了提高,且鲁棒性较好。

关键词: 卷积神经网络, 递归神经网络, 长短期记忆, 脑电信号, 分类识别

Abstract: Due to the low recognition rate of traditional EEG signal classification methods, and the exponentially decreasing recognition rate with the increase of the EEG signal categories, a multi-layered Convolution Bidirectional LSTM recursive neural network(CBLSTM) classification model is designed, aiming for the combination of temporal and spatial characteristics of EEG signals. This model extracts the frequency domain features of EEG sequences by using the multi-layer convolution neural network. It extracts the time domain features of EEG signals by the bidirectional LSTM. EEG signal sequences are input into this model frame by frame. And finally, the classification result can be output. The feasibility of this method is verified by comparative study. Experiments show that the average recognition rate of this classification model is improved , and the robustness is better.

Key words: convolutional neural network, recurrent neural networks, long-short term memory, Electroencephalography(EEG), classification