基于连续卷积神经网络的3D脑电图情绪识别  

Emotion recognition in 3D EEG based on continuous convolutional neural network

作　　者：李文书[1] 王浩 Li Wenshu;Wang Hao(Laboratory of Intelligent Identification and Systems,School of Computer Science,Zhejiang Sci-Tech University,Hangzhou,Zhejiang 310018,China)

机构地区：[1]浙江理工大学计算机学院智能识别与系统实验室,浙江杭州310018

出　　处：《计算机时代》2023年第9期96-100,105,共6页Computer Era

基　　金：国家自然科学基金(31771224,61603228);国家科技部重点研发计划重点专项课题(2018YFB1004901);浙江省自然科学基金(LY17C090011,LGF19FO20009)。

摘　　要：作为人类大脑神经活动的重要指标,脑电(EEG)信号能更深层次地反映人的情绪状况,但当前大多数深度学习模型通过EEG识别情绪不够精确。本文提出基于连续卷积神经网络的3D脑电图情绪识别模型,采用提取DE和PSD特征时得到的等价矩阵所构造的三维脑电图立方体作为输入,舍弃卷积神经网络中的池化层。在DEAP数据集上进行实验,Arousal准确率为95.77%,Valence准确率为93.52%,表明该方法非常适用于情绪识别,能给后续相关研究提供参考。As an important indicator of human brain neural activity,the electroencephalography(EEG)signal can reflect human emotional condition at a deeper level,but most current deep learning models are not accurate enough to identify emotions by EEG.A 3D EEG emotion recognition model based on continuous convolutional neural network is proposed,which uses the 3D EEG cube constructed from the equivalence matrix obtained when extracting DE and PSD features as input and discards the pooling layer in the convolutional neural network.Experiments are conducted on the DEAP dataset with an accuracy of 95.77%for Arousal and 93.52%for Valence,which shows that the method is suitable for emotion recognition tasks and can provide references for subsequent related research.

关 键 词：脑电情绪识别 连续卷积神经网络 深度学习 DEAP数据集 3D脑电图 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]