基于MAResnet的脑电情感识别研究

Research on EEG emotion recognition based on MAResnet

情感是大脑活动的一种表现形式,与心理活动和日常生活密切相关。利用脑电情感数据库并依据心理效价和唤醒度情感划分模型,对压力、平静、轻松、沮丧和快乐5种情感进行研究分析。针对脑电信号时空特征结合的特点,以深度学习中的残差神经网络为基础,提出基于多尺度注意力残差网络(MAResnet)的脑电情感信号分类模型。通过在传统的残差学习模块中加入注意力机制并在同一空间位置并联使用不同尺寸的卷积核,从而对脑电情感信号进行了多尺度特征提取,并对神经网络通过残差学习来避免网络退化。实验结果表明,改进后的多尺度注意力残差网络的分类精度为85.2%,较传统残差网络的分类精度提升了17.7%,较已有相似研究如应用SVM、KNN等方法在分类类型和识别精度上都有显著提升,证明该方法的有效性。

Emotion is a manifestation of brain activity,which is closely related to psychological activity and daily life.Using the publicly available EEG emotional database on the Internet and the emotional classification model based on psychological valence and arousal degree,the five emotions of stress,calmness,relaxation,depression and happiness are studied and analyzed.Aiming at the characteristics of the temporal and spatial feature combination of EEG signals,based on the residual neural network in deep learning,an EEG emotional signal classification model based on multi-scale attention residual networks(MAResnet)is proposed.Through adding attention mechanism into the traditional residual learning module and using the convolution kernels with different sizes in parallel in the same space,the multi-scale features of EEG emotional signals are extracted,and residual learning is performed on the neural network to avoid network degradation.The experiment results show that the classification accuracy of the improved multi-scale attention residual network is 85.2%,which is 17.7%higher than that of the traditional residual network.Compared with the existing similar research such as SVM,KNN and other methods,the classification type and recognition accuracy are significantly improved,which proves the effectiveness of the method.