EEG control intention decoding based on improved discriminant analysis

Abstract

Abstract Decoding the intent of electroencephalographic (EEG) signals is a crucial topic in brain-computer interface research. As a classical multivariate statistical method, discriminant analysis is widely used in EEG-based intent decoding. The core prin-ciple entails building a discriminant model with established observation indices as training samples, enabling the discrimi-nation and classification of unattributed samples based on this model. In the process of deciding the discriminant rules, typical discriminant analysis methods are efficient and simple, but they rely on two traditional estimators leading to the sample mean and the sample scatter matrix, which implies that they lack robustness. This study examines four discriminant analysis methods including linear discriminant analysis, quadratic discriminant analysis, regularized discriminant analysis, general-ized discriminant analysis. The study further introduces robust discriminant analysis, investigating the classifier performance when robustness is enhanced in the estimation of mean vectors and covariance matrices. Research findings indicate that the proposed robust discriminant analysis classifier exhibits superior classification accuracy and enhanced robustness.