Continuous Convolutional Neural Network with 3D Input for EEG-Based Emotion Recognition

Automatic emotion recognition based on EEG is an important issue in Brain-Computer Interface (BCI) applications. In this paper, baseline signals were taken into account to improve recognition accuracy. Multi-Layer Perceptron (MLP), Decision Tree (DT) and our proposed approach were adopted to verify the effectiveness of baseline signals on classification results. Besides, a 3D representation of EEG segment was proposed to combine features of signals from different frequency bands while preserving spatial information among channels. The continuous convolutional neural network takes the constructed 3D EEG cube as input and makes prediction. Extensive experiments on public DEAP dataset indicate that the proposed method is well suited for emotion recognition tasks after considering the baseline signals. Our comparative experiments also confirmed that higher frequency bands of EEG signals can better characterize emotional states, and that the combination of features of multiple bands can complement each other and further improve the recognition accuracy.