Emotion recognition using time-frequency ridges of EEG signals based on multivariate synchrosqueezing transform.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Emotion recognition from electroencephalogram (EEG) signals is crucial for understanding human affective states.
• Extracting informative features from multi-channel EEG with low computational cost remains a challenge.
• Time-frequency (TF) analysis offers potential for detailed signal component characterization.

Purpose of the Study:

• To investigate the feasibility of TF ridges estimation for multi-channel EEG-based emotion recognition.
• To develop a low-computational-cost method for extracting informative components from EEG signals.
• To enhance the accuracy of valence and arousal recognition using advanced signal processing techniques.

Main Methods:

• Utilized multivariate synchrosqueezing transform (MSST) for well-localized TF representation of multi-channel EEG signals.
• Employed TF ridges estimation to identify maximum-energy components, extracting instantaneous frequency and amplitude.
• Computed statistical properties (mean, variance) of five maximum-energy TF ridges as a 20-dimensional feature vector for machine learning.

Main Results:

• Successfully compressed multi-channel EEG component information into a low-dimensional feature space.
• Achieved high recognition rates on the DEAP dataset: up to 71.55% for arousal and 70.02% for valence.
• Demonstrated the effectiveness of TF ridges statistical values as features for emotion recognition.

Conclusions:

• TF ridges estimation using MSST provides an effective and computationally efficient method for EEG-based emotion recognition.
• The proposed feature extraction technique successfully captures and compresses essential information from multi-channel EEG.
• This approach offers a promising direction for developing practical and accurate affective computing systems.