基于4D-CNNLSTM的动态脑网络情绪识别

Dynamic Brain Network Emotion Recognition Based on 4D-CNLSTM

脑电信号是一种时变的非线性空间离散信号,为了反映大脑各区域之间的信息传递与空间关系,提出了一种基于四维特征和卷积长短时记忆网络的情绪识别方法(4D-CNNLSTM)。该方法将原始的脑电信号转换为二维平面-频率-时间的4D特征,随后使用卷积和长短时记忆网络学习特征的空间、频率和时间信息,最后使用SoftMax进行分类。实验结果表明,该方法具有良好的性能,情绪识别的准确率最高达到91.87%。

The EEG signal is a time-varying,nonlinear spatial discrete signal.In order to reflect the information transfer and spatial relationships among brain regions,an emotion recognition method(4D-CNLSTM)based on four-dimensional features and convolutional long-short time memory networks is proposed.The method converts raw EEG signals into 2D plane-frequency-time 4D features,then uses convolution and long-short time memory networks to learn the spatial,frequency and temporal information of the features,and finally uses SoftMax for classification.The experimental results indicate that the method has good performance,with an accuracy of up to 91.87%for emotion recognition.