Abstract:

In recent years, more and more researchers pay attention to emotional recognition by analyzing EEG signals. In order to enrich the feature representation and obtain a higher classification accuracy of emotional recognition, this paper applies MFCC, a speech cepstrum feature, to EEG signals. After wavelet transform of EEG signal, MFCC feature and EEG feature are extracted and fused. Affective classification and recognition are carried out by using the characteristics of ResNet18. The experimental results show that, compared with traditional EEG features, the addition of MFCC features improves the recognition accuracy of emotional dimension Arousal and Valence by 6% and 4%, respectively, reaching 86.01% and 85.46%, thus improving the recognition accuracy of emotional recognition.

Key words: EEG signal, Mel-scale Frequency Cepstral Coefficients（MFCC）, feature fusion, deep residual network

摘要：

近年来，通过分析脑电图（EEG）信号来实现情感识别的课题越来越被研究者所重视。为了丰富特征的表示能力，获得更高的情感识别分类准确率，尝试将语音信号特征梅尔频率倒谱系数MFCC应用于脑电信号。在对EEG信号小波变换的基础上将提取得到的MFCC特征与EEG特征相互融合，通过利用深度残差网络（ResNet18）的特性进行情感分类识别。实验结果表明，比起传统的单一利用EEG特征，添加了MFCC特征使得情感维度Arousal和Valence两者的识别准确率分别提升了6%和4%，达到了86.01%和85.46%，从而提升了情感的识别准确度。

关键词: 脑电信号, 梅尔倒谱系数（MFCC）, 特征融合, 深度残差网络