基于三维特征矩阵和冲压激励网络的多通道脑电情感识别  

Emotion recognition from multi-channel EEG data through three-dimensional feature matrix and squeeze-and-excitation networks

作　　者：晁浩[1] 曹益鸣 刘永利[1] CHAO Hao;CAO Yi-ming;LIU Yong-li(School of Computer Science and Technology,Henan Polytechnic University,Jiaozuo 454000,China)

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454000

出　　处：《控制与决策》2023年第12期3427-3435,共9页Control and Decision

基　　金：国家自然科学基金项目(61872126);河南省高等学校重点科研计划项目(19A520004);河南省高校基本科研业务费专项资金项目(NSFRF1616)。

摘　　要：提出一种基于冲压激励网络的情感状态识别方法.首先,从不同通道的脑电信号中提取时域特征,并根据电极通道的相对位置构造三维特征矩阵;然后,将冲压激励块与三维卷积神经网络相结合构建冲压激励网络进行高层抽象特征提取;最后,使用全连接层进行情感状态分类.实验在DEAP数据集上开展,实验结果表明,冲压激励网络在利用脑电信号中的时域显著性信息和电极空间位置信息的基础上,可自适应地纠正特征的注意力,优化每个特征的权重并强化重要特征,同时利用不同特征的互补信息来提高识别精度;此外,冲压激励网络的挤压操作可获取输入数据的全局信息,具有较快的收敛速度.An emotion state recognition method based on squeeze-and-excitation networks is proposed.Firstly,time-domain features are extracted from electroencephalogram(EEG) signals of different channels,and a three-dimensional feature matrix is constructed according to the relative position of electrode channels.Then,the squeeze-and-excitation networks are constructed by combining the squeeze-and-excitation block with the 3D convolutional neural network for high-level abstract feature extraction.Finally,fully connected layers are used for emotional state classification.The experiment is carried out on the DEAP data set.The experimental results show that squeeze-and-excitation networks can adaptively correct the attention of features and optimize the weight of each feature based on the time-domain saliency information and electrode spatial position information in the EEG signal.Meanwhile,the squeeze-and-excitation networks can also strengthen important features and improve the recognition accuracy by using the complementary information of different features.In addition,the squeeze operation of the squeeze-and-excitation networks can obtain the global information of the input data and have faster convergence speed.

关 键 词：情感识别 多通道脑电 冲压激励网络 三维特征矩阵 注意力 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]