A Novel Approach to Feature Extraction in MI - Based BCI Systems

Brain-computer interfaces based on motor imagery (MI-BCI) enable communication between the brain and the outside world. It accomplishes this by capturing and examining the EEG data generated when a person visualizes moving a certain limb. Based on this investigation; to enhance categorization precision, we enhance the attribute extraction process. Once the EEG signals have been pre-processed and deconstructed into distinct bands of frequencies, we apply similar spatial algorithms for every sub-band. After that, we use Blind Source Separation (BSS) and spatial filter techniques to extract attributes directly from the sources of the brain instead of only from channels. Ultimately, by choosing the most relevant attributes as inputs for the categorization stage, we distinguish between the envisioned movements. To assess our recommended approach, we utilized data set IVa from the BCI competition III. The outcomes from diverse experiments indicate that the system achieves an average accuracy of 98.8%, with 100% sensitivity across all subjects. These results illustrate that the suggested approach effectively improves MI-based BCI.