EEG-Based Emotion Recognition using 3D Convolutional Neural Networks

Emotion recognition is a crucial problem in Human-Computer Interaction (HCI). Various techniques were applied to enhance the robustness of the emotion recognition systems using electroencephalogram (EEG) signals especially the problem of spatiotemporal features learning. In this paper, a novel EEG-based emotion recognition approach is proposed. In this approach, the use of the 3-Dimensional Convolutional Neural Networks (3D-CNN) is investigated using a multi-channel EEG data for emotion recognition. A data augmentation phase is developed to enhance the performance of the proposed 3D-CNN approach. And, a 3D data representation is formulated from the multi-channel EEG signals, which is used as data input for the proposed 3D-CNN model. Extensive experimental works are conducted using the DEAP (Dataset of Emotion Analysis using the EEG and Physiological and Video Signals) data. It is found that the proposed method is able to achieve recognition accuracies 87.44% and 88.49% for valence and arousal classes respectively, which is outperforming the state of the art methods.