| 杨晓同,易灿南,蒋复量,胡鸿,张勉,李昊宇,吴文.基于CNN-BiLSTM的脑电情绪分类模型及验证[J].南华大学学报(自然科学版),2023,(6):41~49.[YANG Xiaotong,YI Cannan,JIANG Fuliang,HU Hong,ZHANG Mian,LI Haoyu,WU Wen.EEG Emotion Classification Model and Validation Based on CNN-BiLSTM[J].Journal of University of South China(Science and Technology),2023,(6):41~49.] |
| 基于CNN-BiLSTM的脑电情绪分类模型及验证 |
| EEG Emotion Classification Model and Validation Based on CNN-BiLSTM |
| 投稿时间：2023-09-01 |
| DOI： |
| 中文关键词: 情绪分类 脑电 神经网络 特征融合 |
| 英文关键词:emotion classification electroencephalogram(EEG) neural network feature fusion |
| 基金项目:湖南工学院省级应用特色学科开放基金(KFB22028;KFB22029);湖南省教育厅项目(21C0811);国家级大学生创新创业训练计划项目(2022211528009) |
|
| 摘要点击次数: 50 |
| 全文下载次数: 31 |
| 中文摘要: |
| 情绪在人们的思考、行为和交流方式中起着重要作用。为提高脑电信号的情绪识别准确率,充分利用脑电信号的频率、空间和时间维度上的信息,提出一种基于CNN-BiLSTM(convolutional neural networks-bidrectional long short term memory)的脑电情绪分类神经网络模型。该模型由卷积神经网络和多层特征融合的双向长短时神经网络构成,卷积神经网络用于学习脑电信号的频率和空间特征,双向长短时神经网络则从卷积神经网络的输出中挖掘脑电切片之间的时序信息。借助离散情绪模型的SEED(sjtu emotion eeg dataset)数据集和连续情绪模型的DEAP(database for emotion analysis using physiological signals)数据集来进行情绪分类实验。实验结果表明,在SEED和DEAP两个数据集上,CNN-BiLSTM模型均取得了目前最好的情绪分类性能。此外,该模型的时序信息挖掘模块性能优于单层长短时神经网络,能够学习更多的时序信息。 |
| 英文摘要: |
| Emotions play an important role in people's thinking, behavior and communication styles. In order to improve the accuracy of emotion recognition from EEG (electroencephalogram) signals and make full use of the information in the frequency, spatial and temporal dimensions of EEG signals, a CNN-BiLSTM(convolutional neural networks-bidrectional long short term memory)-based neural network model for EEG emotion classification is proposed. The model consists of a convolutional neural network and a bi-directional long-short-time neural network with multi-layer feature fusion, in which the convolutional neural network is used to learn the frequency and spatial features of the EEG signals, and the bi-directional long-short-time neural network mines the temporal information between EEG slices from the output of the convolutional neural network. Emotion classification experiments are conducted with the help of the SEED dataset for the discrete emotion model and the DEAP dataset for the continuous emotion model. The experimental results show that the CNN-BiLSTM model achieves the best emotion classification performance so far on both SEED and DEAP datasets. In addition, the temporal information mining module of the model outperforms single-layer LSTM(long short term memory,LSTM) and is able to learn more temporal information. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|