Fusion of Hadamard and Riemannian Space Features for Motor Imagery EEG Classification

Résumé fourni par la source

Motor Imagery (MI) has received particular attention due to its ability to allow voluntary control without actual movement among various Brain-Computer Interface (BCI) paradigms, particularly in applications aimed at assisting individuals with motor disabilities. Motor imagery classification poses considerable difficulties because of its complexity and nonstationary nature of EEG signals, which are highly subject-dependent and influenced by inter-session variability, as well as demographic factors such as age and gender. Additionally, EEG signals are highly susceptible to both internal physiological noise and external environmental disturbances, further complicating accurate classification. To address these challenges, we propose a novel feature extraction and classification pipeline that integrates Fisher Geodesic Discriminant Analysis (FGDA), tangent space mapping (TMS), and the Hadamard transformation (HT) to enhance MI EEG decoding. For training and testing, the BCI Competition IV-2a dataset was utilized, achieving a classification accuracy of 92.30% for subject-dependent and 71.67% for subject-independent classification. This represents an average improvement of 11.0% and 10.43% for subject-dependent and subject-independent respectively over previously reported state-of-the-art methods. Additionally, it demonstrated superior average accuracy across subjects, confirming its robustness and effectiveness in motor imagery EEG classification tasks.

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