Classifying EEG-based Upper Limb Motor Imagery Tasks for Brain-Robot Interface-based System Development

摘要

Cognitive robotics and engineering strive to develop brain-robot interfaces (BRI) that enable effective collaboration between robots and humans. The continued advancements in neuroscience, robotics, and machine learning are projected to broaden the scope of BRI applications, ultimately improving human-robot cooperation. A major area of focus for BRI-based system development involves mapping brain activities to control actions through motor imagery (MI) based technologies such as electroencephalogram (EEG). In this paper, a novel technique for classifying different tasks involving MI of upper limb movements from EEG signals is presented. To emphasize the inter-related information within various bands of EEG signals, ratio of band power (RBP)-based features are proposed for categorizing different upper limb-based MI tasks. The combination of these proposed features and an optimized KNN classifier produced remarkable accuracy in classifying diverse MI tasks from EEG signals. This approach was compared with recent methods applied to the same dataset, illustrating its superiority in performance. Furthermore, a comparison between the proposed RBP-based features and conventional band power-based features underscored the effectiveness of RBP-based attributes for classifying MI-related tasks. The improved effectiveness of the proposed approach could lead to progress in BRI-based systems development across a range of applications.