Emotion Recognition of EEG Signals Based on 3D Parallel Multi-Field Convolutional Neural Network
The use of EEG signals to identify emotional states has become a popular research topic.Existing emotion recognition methods usually extract two-dimensional information as samples,but ignore the spatial information that containing important features of different re-gions of the brain.A three-dimensional parallel multi-field convolutional neural network(TPMCNN)based on the layout of EEG channels and the frequency-related features in the original EEG signal is proposed to address this problem.Firstly,the original EEG signal is divid-ed into multiple frequency bands,and the differential entropy(DE)features of each band are extracted.Then the data are transformed into a 3D feature matrix according to the location of the electrode sensors.Finally the resulting matrix is processed by using the TPMCNN.The experimental results showe that the 3D feature matrix constructed using differential entropy features of different frequency bands can effec-tively extract features related to emotion recognition from multi-channel EEG signals,and the proposed parallel multi-field convolutional neural network can fully exploit the advantages of deep learning.The experiments are performed on the publicly available dataset of DEAP for dichotomous classification,and the values of accuracy reach 97.31%and 96.72%for arousal and valence respectively,and 97.17%for four-classification,confirming the superior performance of the proposed method for EEG signal emotion recognition.