Electroencephalogram Access for Emotion Recognition Based on a Deep Hybrid Network

Area of Science:

• Human-Computer Interaction
• Neuroscience
• Machine Learning

Background:

• Electroencephalogram (EEG) signals offer a viable pathway for automatic emotion recognition in human-computer interaction (HCI).
• Existing methods for EEG-based emotion recognition require enhancement in accuracy and robustness.
• Deep learning models present opportunities for advanced analysis of complex EEG data.

Purpose of the Study:

• To propose and evaluate a novel deep hybrid network for improved EEG-based emotion recognition.
• To enhance the accuracy of identifying human emotional states (arousal, valence, dominance) using EEG data.
• To validate the effectiveness of the proposed method against state-of-the-art techniques.

Main Methods:

• Collected EEG data was decomposed into four frequency bands.
• Multiscale Sample Entropy (MSE) features were extracted from each frequency band.
• A deep hybrid network, combining a Convolutional Neural Network (CNN) and Hidden Markov Models (HMMs), processed 3D MSE feature matrices.
• HMM classifiers replaced traditional artificial neural network classifiers within the CNN framework.

Main Results:

• The proposed method achieved an average accuracy of 79.77% for arousal, 83.09% for valence, and 81.83% for dominance on the DEAP dataset.
• Significant improvements were observed compared to existing methods, with a 0.99% increase in valence accuracy and a 14.58% increase in dominance accuracy.
• The HMM-based classification demonstrated superior performance in emotion recognition tasks.

Conclusions:

• The deep hybrid network integrating CNN and HMMs effectively enhances EEG-based emotion recognition accuracy.
• The multiscale sample entropy feature extraction combined with the hybrid network architecture proves effective for capturing emotional nuances.
• This approach represents a significant advancement in developing more perceptive and responsive human-computer interaction systems.