Classification Methods for EEG Patterns of Imaginary Movements

The review focuses on the most promising methods for classifying EEG signals non-invasive BCIs and theoretical approaches successful classification of patterns. paper provides an overview articles using Riemannian geometry, deep learning various options preprocessing "clustering" signals, example, common-spatial pattern (CSP). Among other approaches, pre-processing CSP is often used, both offline online. combination CSP, linear discriminant analysis, support vector machine neural network (BPNN) made it possible to achieve 91% accuracy binary with exoskeleton control as a feedback. There very little work use geometry online best achieved so far problem 69.3% in work. At same time, testing, average percentage correct considered – 77.5 ± 5.8%, networks 81.7 4.7%, 90.2 6.6%. Due nonlinear transformations, geometry-based complex provide higher better extract useful information from raw recordings rather than transformation. However, real-time setup, not only important, but also minimum time delay. Therefore, transformation delay less 500 ms may be future advantage.