An innovative brain-inspired neuronal competition model for electroencephalography emotion recognition

Area of Science:

• Neuroscience
• Computer Science
• Digital Health

Background:

• Emotion recognition is crucial for human-computer interaction and clinical neuropsychology.
• Current deep learning models for emotion recognition lack interpretability, hindering insights into brain activity.

Purpose of the Study:

• To propose an innovative brain-inspired neuronal competition model for emotion recognition from electroencephalography (EEG).
• To enhance the interpretability of emotion recognition models by simulating biological neural processes.

Main Methods:

• Feature extraction using discrete wavelet transform to obtain five brain rhythms (delta, theta, alpha, beta, gamma).
• A neuronal competition model simulating excitatory and inhibitory dynamics.
• Rhythm attention encoder and leaky integrate-and-fire model for neuronal decision-making.
• Emotion recognition based on a winner-take-all principle.

Main Results:

• Achieved average accuracies of 86.64-92.70% on the DEAP dataset.
• Attained an accuracy of 81.89% on the SEED dataset.
• Reached accuracies of 80.38-82.70% on the DREAMER dataset.
• Demonstrated that different brain rhythms have distinct characteristics in emotion recognition tasks.

Conclusions:

• Presents a biologically plausible paradigm for EEG-based emotion recognition.
• Provides a foundation for understanding the neurophysiological basis of brain activity.
• Aims to develop reliable tools for healthcare applications like mental health monitoring and depression detection.