Multi-Modal Emotion Recognition Using EEG and Eye Tracking Features.

Multi-modal emotion recognition from various human physiological indicators has emerged as a large topic of interest, including the use of EEG, ECG, GSR and Eye Tracking features. This work introduced a simple CNN based multi-modal EEG and Eye Tracking emotion recognition model for the SEED V dataset. In contrast to other works on the SEED V dataset, different Differential Entropy time windows were tested for EEG feature extraction. EEG signals were arranged in a 2D image format to preserve spatial relationships between electrode placements on patients during the trials. The proposed model with a 1 second processing window for EEG features achieved state of the art results in Leave One Subject Out Validation, with a mean accuracy of 0.935 ± 0.038 on the SEED V dataset. A noticeable improvement was noted over the same multi-modal model using a 4 second processing window for EEG features, highlighting the importance of smaller time windows for EEG feature processing in emotion recognition problems.