MI recognition by subject specific localised frequency fusion

EEG-based motor imagery (MI) discrimination is widely utilised in real-life applications, such as brain-computer interfaces (BCIs), due to its non-invasive and comparatively cost-effective nature. However, traditional BCI systems typically rely on a uniform, broad frequency band or a standardised set of sub-bands to extract features. Consequently, they fail to account for distinct physiological variability across subjects, leading to significant classification performance degradation. To address this limitation, we propose a novel framework named subject-specific localised frequency fusion (SSLFF). The proposed method systematically searches for and selects optimal frequency sub-bands , while dynamically merging complementary spectral bands. In addition to this subject-specific frequency localisation, the framework integrates a time-localised feature extraction process. Unlike traditional approaches, the proposed method can operate effectively over a broader master frequency band without experiencing performance degradation, as it automatically isolates the optimal spectral parameters for each subject. Overall, the proposed framework achieves a statistically significant improvement in classification accuracy compared to alternative traditional methods, while maintaining exceptional robustness against variations in system parameters. In short, SSLFF try to address the fact that the learning of human brain varies from person to person and it incorporates subject specific tuning not only in the spatial filter and classifier, but also in frequency band selection and localised feature extraction, leading to a superior classification performance.