Improved filter bank common spatial pattern algorithm based on the sparrow search algorithm

Le résumé fourni par la source

Introduction: The application of motor imagery in human-computer interaction and rehabilitative medicine has attracted growing attention due to recent advances in brain-computer interface technologies. However, traditional EEG decoding paradigms based on fixed frequency-band segmentation often exhibit limited performance because they fail to capture individual variability in brain rhythms. Methods: This work proposes an adaptive method that integrates the sparrow search algorithm (SSA) with Filter Bank Common Spatial Pattern (FBCSP) to optimize sub-band segmentation for motor imagery EEG decoding. SSA adaptively searches for optimal sub-band boundaries, enabling individualized frequency-band selection. Results: Experiments on the BCI Competition IV 2a dataset under a cross-session evaluation protocol (training on session T, testing on session E) demonstrated that SSA-FBCSP effectively improves frequency-band adaptability. The SSA-FBCSP approach was further combined with Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and k-Nearest Neighbor (KNN) classifiers to evaluate the influence of different downstream classifiers. Conclusion: Among them, SSA-FBCSP-LDA achieved the best performance, outperforming the conventional uniform sub-band approach by 21.76% and reaching an average accuracy of 89.92%. The adaptively selected sub-bands closely matched the ERD/ERS distribution, confirming the method's effectiveness in frequency-band optimization. Compared with recent deep-learning-based MI-EEG models, the proposed technique offers a balance of accuracy, interpretability, and computational efficiency, providing a promising direction for personalized brain-computer interface systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.