A novel channel selection scheme for olfactory EEG signal classification on Riemannian manifolds

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Olfactory electroencephalogram (EEG) signal classification has diverse applications but faces challenges with high channel counts, leading to redundancy and computational load.
• Effective channel selection is crucial for improving the precision and efficiency of olfactory EEG signal classification.

Purpose of the Study:

• To propose and evaluate a Multi-Strategy Fusion Binary Harmony Search (MFBHS) algorithm for optimal channel selection in olfactory EEG signals.
• To reduce the number of EEG channels required for accurate classification while minimizing information redundancy and computational complexity.

Main Methods:

• Developed the MFBHS algorithm by integrating opposition-based learning, adaptive parameter control, and bitwise operations into a binary harmony search framework.
• Implemented channel selection directly on the covariance matrix of EEG signals, evaluating subsets using classification accuracy and channel count.
• Combined MFBHS with a Riemannian geometry classification framework for olfactory EEG signal analysis.

Main Results:

• MFBHS successfully minimized the number of selected EEG channels while achieving classification accuracy comparable to using all channels across multiple protocols and datasets.
• The selected channels demonstrated subject-independent generalization, performing well on untrained subjects without significant accuracy loss.
• MFBHS outperformed existing state-of-the-art channel selection algorithms in terms of efficiency and accuracy.

Conclusions:

• The MFBHS algorithm provides a practical and effective solution for optimizing EEG channel usage in olfactory recognition tasks.
• This approach enhances the feasibility of applying olfactory EEG analysis in real-world applications by reducing data requirements and improving computational efficiency.