MEMD-HHT based Emotion Detection from EEG using 3D CNN.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Emotion detection from electroencephalography (EEG) signals is crucial for human-computer interaction.
• Extracting robust features from complex EEG data remains a challenge.

Purpose of the Study:

• To develop an advanced feature extraction and classification framework for accurate emotion detection using multichannel EEG.
• To leverage Multivariate Empirical Mode Decomposition (MEMD) for identifying emotion-related signal variations.

Main Methods:

• Multichannel EEG data was processed using MEMD to extract scale-aligned intrinsic mode functions (IMFs).
• High-frequency IMFs were selected as significant features, and the Marginal Hilbert Spectrum (MHS) was computed.
• A 3D Convolutional Neural Network (CNN) was employed for emotion classification using spatio-temporal-spectral features derived from MHS.

Main Results:

• The proposed method achieved high accuracy in binary classification of valence (89.25%) and arousal (86.23%) levels.
• The approach demonstrated superior performance compared to existing methods using 2D CNNs or conventional features.
• The study confirmed the significance of high-oscillatory IMFs and MHS for emotion detection.

Conclusions:

• MEMD provides effective features for EEG-based emotion recognition.
• The 3D CNN model integrated with MHS features offers a powerful tool for emotion detection.
• This research advances the state-of-the-art in affective computing through improved EEG signal analysis.