This work presents the research report titled "Brain-Machine Interface: an optimal approach via Riemannian distance,"developed from September 1, 2023, to August 31, 2024, during the execution of the research project named "Optimization techniques applied to Brain-Machine Interface,"funded by the Amazon Foundation for Research Support, under the supervision of Professor Dr. Cleison Daniel Silva. This work was prepared following resolution nº1/2024 of the Faculty of Electrical Engineering - CAMTUC, which regulates the terms of the flexibilization of the Final Graduation Project in IN nº5/2023
of PROEG-UFPA. Brain-Machine Interface (BMI) systems are technologies capable of establishing communication between the human brain and external devices through neural signals, which can be collected via neuroimaging techniques such as electroencephalography (EEG), processed, and converted into commands. The research study focuses on improving classification performance in motor imagery-based BMI systems using the Minimum Distance to Riemannian Mean (MDRM) method through the Minimum Distance to Mean (MDM) classification algorithm for extracting discriminative information from EEG signals represented by positive definite symmetric covariance matrices defined by sub-bands, forming a normalized representation of the EEG signals that are fed into the Support
Vector Machine (SVM) classification algorithm. The hyperparameters related to the frequency band of interest, the number of sub-bands, and classifier parameters are adjusted through Bayesian Optimization to address inter- and intra-subject characteristics, allowing for individual adjustments. The results obtained from a public dataset showed significant improvements compared to a previously proposed method. The classifier’s accuracy was used for comparison, serving as the basis for discussions and conclusions of the research.
|
