Optimized Projection and Fisher Discriminative Dictionary Learning for EEG Emotion Recognition

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalogram (EEG)-based emotion recognition (ER) is crucial for human-machine interaction.
• Current methods often struggle to leverage band-specific EEG information effectively.
• Existing techniques typically concatenate frequency band features, limiting the utilization of unique band characteristics.

Purpose of the Study:

• To propose an optimized projection and Fisher discriminative dictionary learning (OPFDDL) model for EEG-based ER.
• To efficiently exploit specific discriminative information from individual EEG frequency bands.
• To preserve shared discriminative information across multiple frequency bands.

Main Methods:

• Subspace projection technology is used to project EEG signals from all frequency bands into a shared subspace.
• A shared dictionary is learned within the projection subspace.
• The Fisher discrimination criterion is applied to dictionary atoms to optimize sparse reconstruction errors.
• An alternating optimization algorithm is developed for learning the projection matrix and dictionary.

Main Results:

• The OPFDDL model demonstrated remarkable results on two EEG-based ER datasets.
• The model effectively utilizes band-specific discriminative information.
• It successfully preserves shared discriminative information among different frequency bands.

Conclusions:

• The proposed OPFDDL model offers an effective approach for EEG-based emotion recognition.
• This method enhances the utilization of multi-band EEG features.
• The findings highlight the model's potential in advancing brain-computer interface applications.