An Innovative Multi-Model Neural Network Approach for Feature Selection in Emotion Recognition Using Deep Feature...

Area of Science:

• Neuroscience
• Machine Learning
• Human-Computer Interaction

Background:

• Emotional awareness perception is crucial for natural human-machine interaction.
• Electroencephalography (EEG) is a key technology for measuring user emotional states.
• High-dimensional EEG features lead to significant computational costs.

Purpose of the Study:

• To introduce Deep Feature Clustering (DFC) for efficient EEG feature selection.
• To reduce computational cost and training time in emotion recognition systems.
• To enhance the accuracy and competitiveness of emotion recognition methods.

Main Methods:

• EEG signal decomposition using Empirical Mode Decomposition (EMD).
• Feature extraction via Analytic Wavelet Transform (AWT) and pre-trained Deep Neural Networks (DNNs).
• Dimensionality reduction and feature selection using DFC and differential entropy-based channel selection.

Main Results:

• DFC effectively selects high-quality attributes, omitting unusable ones.
• The proposed method significantly shortens network training time.
• Improved emotion recognition performance on SEED, DEAP, and MAHNOB datasets.

Conclusions:

• The DFC method offers a competitive and efficient approach to EEG-based emotion recognition.
• This technique enhances recognition performance with reduced processing time.
• The findings pave the way for more sophisticated and responsive human-computer interaction systems.