深度残差网络结合长短期记忆的脑电波情感识别

EEG Emotion Recognition Based on Deep Residual Network Combined with LSTM

针对传统情感识别方法中脑电信号特征提取与辨识困难的问题,本文提出一种基于深层残差网络和长短时记忆网络的情感识别方法。首先,将DEAP生理数据信号进行连续小波变换得到相应的时频谱图;然后对时频谱图进行灰度化和归一化,再将灰度图降维至适当大小;最后将压缩后的时频谱图作为深层残差网络的输入,将DRN学习到的顶层特征进行向量化,并输入长短时记忆网络网络实现情感识别。实验结果表明:提出的CWT-DRN-LSTM模型情感识别准确率达99.23%,标准差仅0.27,相比于其它组合模型在识别准确率方面具有较大优势。

Aiming at the difficulty of feature extraction and recognition of emotion states,this paper proposes a recognition method based on deep residual network and long short-term memory.Firstly,the DEAP physiological data were transformed by continuous wavelet transform to obtain the corresponding time frequency images.Then,the time frequency images were grayed and normalized,with its dimension reduced to the appropriate size.The compressed time frequency images were fed into deep residual network,and the top features learned by DRN were vector quantized.Finally,the LSTM network was utilized for emotion recognition.The experimental results show that the emotion recognition accuracy of the proposed CWT-DRN-LSTM model is 99.23%,and the standard deviation is only 0.27,which has great advantages in recognition accuracy compared with other combined models.