Emotion Recognition Tool for Brain-Computer Interface

This work aims to introduce a multimodal framework for emotion recognition. It permits to acquire the signals from electrocardiogram and electroencephalogram devices (e.g., OpenBCI Ultracortex Mark IV headset) and processing them by Deep Learning based techniques for data augmentation and emotion recognition. In this work we focus on practical experiments, for obtaining a pretrained model, by using the SEED-IV dataset with a Variational Autoencoder for data augmentation, and both Continuous Convolutional Neural Network and Graph Isomorphism Network for emotion classification. By continuous Convolutional Neural Network a classification accuracy of 97.28%, by using filtering techniques (moving average and Linear Dynamical System) applied to the Power Spectrum Density, is achieved. On the other hand, Graph Isomorphism Network, suitable for graph-based network classification, achieved an accuracy of 98.4% on the SEED-IV dataset using Differential Entropy features filtered by LDS.