基于深度森林的脑电情绪识别研究

Research on EEG Emotion Recognition Based on Deep Forest

为了改善传统脑电情绪识别方法需要对脑电信号进行深入了解,且需要人工提取相关特征的缺点,基于深度森林的表征学习能力对脑电样本的时域与频域数据进行自动特征提取,并融合32通道脑电信号的时域特征向量和频域特征向量,通过级联森林对特征作进一步学习。实验结果表明,该方法对效价二分类预测的准确率达到68.4%,查准率达到66.3%,查全率达到89.9%,F1分数达到76.3%;对唤醒度二分类预测的准确率达到68.2%,查准率达到65.8%,查全率达到91.2%,F1分数达到76.4%。通过与DEAP数据集使用EEG信号给出的二分类实验结果进行对比,基于深度森林的脑电情绪识别方法对未知样本的识别准确率高于DEAP的结果。

In order to improve the traditional EEG emotion recognition method,the EEG signals need to be deeply understood,and the shortcoming of relevant features need to be extracted manually.Based on the representation learning ability of deep forest,the time-domain data and frequency-domain data of EEG samples were automatically extracted.Then,the time-domain feature vectors and frequency-domain feature vectors of 32-channel EEG signals were integrated.The features were further learned through the cascade forest.Experimental results show that the binary classification accuracy rate of valence is 68.4%,the precision rate is 66.3%,the recall rate is 89.9% and the F1 score is 76.3%.The binary classification accuracy rate of arousal reached 68.2%,the precision rate reached 65.8%,the recall rate reached 91.2%,and the F1 score reached 76.4%.EEG recognition based on deep forest can not only extract features automatically,but also identify unknown samples more accurately than the results in DEAP paper.