Variable length particle swarm optimization and multi-feature deep fusion for motor imagery EEG classification

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) offer a novel communication channel between humans and external systems.
• Accurate classification of motor imagery electroencephalogram (EEG) signals is crucial for effective BCI performance.
• Existing algorithms require improvement in feature extraction and classification for enhanced accuracy.

Purpose of the Study:

• To propose a novel algorithm, VLPSO-MFDF, for improved motor imagery EEG signal classification.
• To enhance the accuracy and performance of BCIs through advanced feature fusion and optimization techniques.

Main Methods:

• A novel algorithm, VLPSO-MFDF, combining variable length particle swarm optimization (VLPSO) and multi-feature deep fusion (MFDF) was developed.
• Deep forest layers were reconstructed into classification modules, processing multiple extracted EEG signal features separately.
• VLPSO was employed to optimize weights for probability vectors, continuously refining classification performance.

Main Results:

• The VLPSO-MFDF algorithm demonstrated significantly higher classification accuracy for four types of motor imagery EEG signals compared to traditional deep forest methods.
• The proposed approach successfully fused multi-domain features and corrected prediction discrepancies.
• Experimental validation confirmed the efficacy of the method in improving classifier performance.

Conclusions:

• The VLPSO-MFDF algorithm represents a significant advancement in motor imagery EEG signal classification for BCIs.
• This method offers a robust approach to fusing diverse features and optimizing classification, paving the way for more reliable human-computer interaction.
• The study highlights the potential of advanced optimization and deep learning fusion techniques in BCI development.