基于1D-CNN-LSTM-SVM模型的脑电情绪识别研究

EEG emotion recognition based on 1D-CNN-LSTM-SVM model

针对情绪识别方法存在特征信息提取不完备和分类模型自适应能力差等问题,提出一种基于一维信息的卷积神经网络(convolutional neural networks,CNN)、长短时记忆神经网络(long short-term memory,LSTM)和支持向量机(support vector machine,SVM)组合模型用于脑电信号(electroencephalogram,EEG)情绪分类。首先,将预处理后的脑电信号输入到1D-CNN-LSTM模型中进行深度特征提取;然后,将输出的多通道融合情感特征输入SVM而不是传统的Softmax进行分类;最后,在脑电情感数据集DEAP上进行情感识别验证,即在唤醒度-效价平面对高效价高唤醒度(HVHA)、高效价低唤醒度(HVLA)、低效价高唤醒度(LVHA)和低效价低唤醒度(LVLA)4个情绪区域进行分类。实验结果表明:1D-CNN-LSTM-SVM的平均准确率优于单独使用CNN算法或1D-CNN-LSTM,情绪识别准确率可达98.20%。该组合模型在执行情绪分类任务时具有良好的鲁棒性,验证了文中提出方法的可行性和有效性。

A combination model with one-dimensional convolutional neural networks(CNN),long short-term memory(LSTM)and support vector machine(SVM)is proposed for the electroencephalogram(EEG)emotion classification to address the problems of incomplete feature information extraction and poor adaptive ability of classification models in emotion recognition methods.Firstly,the preprocessed EEG signals were input into CNN and CNN-LSTM models for deep feature extraction,respectively;secondly,the output multi-channel fused emotional features were input into the SVM instead of the traditional Softmax for classification;finally,validation of emotion recognition was performed on the EEG emotion dataset DEAP,categorizing into four emotion regions in terms of arousal and valence:high valence high arousal(HVHA),high valence low arousal(HVLA),low valence high arousal(LVHA)and low valence low arousal(LVLA).The experimental results show that the average accuracy of the 1D-CNN-LSTM-SVM outperforms that of the CNN algorithm or 1CNN-LSTM,with an emotion recognition accuracy of up to 98.20%.The combined model has good robustness in performing the emotion classification task,which verifies the feasibility and effectiveness of the proposed method in the paper.