多通道输入多尺度卷积神经网络脑电情绪识别
EEG Emotion Recognition Based on Multi-channel Input and Multi-scale Convolutional Neural Network
梁椰舷 1李婷1
作者信息
- 1. 西安工程大学计算机科学学院 西安 710048
- 折叠
摘要
针对传统方法提取出的脑电特征单一且不充分的问题,论文在经典卷积神经网络的基础上进行改进,提出了一种具有不同尺度卷积核的卷积神经网络(CNN)模型,并在DEAP数据集上进行了验证.论文实验首先结合每位被试的基线信号进行微分熵特征提取,预处理后的结果在频率维度组合为四通道形式,最后输入具有不同尺度卷积核的CNN模型中进行分类实验,该方法在唤醒度和效价上的分类准确率为97.77%和97.55%.相比于单一尺度CNN模型准确率有了明显的提升,证明了论文网络模型的有效性.
Abstract
Aiming at the problem that the electroencephalogram(EEG)features extracted by traditional methods are single and insufficient,this paper improves the classical convolution neural network,proposes a convolution neural network(CNN)model with different scale convolution kernel,and verifies it on DEAP dataset.In this experiment,the differential entropy(DE)feature is extracted by combining the baseline signal of each subject.The preprocessed results are combined into four channel form in the fre-quency dimension.Finally,the classification experiment is input into the CNN model with different scale convolution kernel.The classification accuracies of this method in arousal and valence are 97.77%and 97.55%.Compared with the single scale CNN model,the accuracies are significantly improved,which proves the effectiveness of the network model in this paper.
关键词
微分熵(DE)/脑电信号(EEG)/卷积神经网络(CNN)/情绪识别Key words
differential entropy(DE)/electroencephalogram(EEG)/convolution neural network(CNN)/emotion recognition引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据