Removing Unrelated Features Based on Linear Dynamical System for Motor-Imagery-Based Brain-Computer Interface

Abstract

Abstract. Common spatial pattern (CSP) is very successful in constructing spatial filters for detecting event-related synchronization and event-related desynchronization. In statistics, a CSP filter can optimally separate the motor-imagery-related components. However, for a single trail, the EEG features extracted after a CSP filter still include features not related to motor imagery. In this study, we introduce a linear dynamical system (LDS) approach to motor-imagery-based brain-computer interface (MI-BCI) to reduce the influence of these unrelated EEG features. This study is conducted on a BCI competition data set, which comprises EEG signals from several subjects performing various movements. Experimental results show that our proposed algorithm with LDS performs better than a traditional algorithm on average. The results reveal a promising direction in the application of LDS-based approach to MI-BCI.

High performance EEG signal classification using classifiability and the Twin SVM

“…Also, as compared to the classification accuracy obtained by the work in [39], our results indicate substantial improvement. For the subject k6b and l1b, the average classification accuracy is reported as 60-70% and 65-78% across tasks respectively.…”

“…We compare our work with these approaches in Sections 4.2 and 4.3. We also compare our work with the results obtained in [39] which uses sub-band filtering approach followed by CSP on the same dataset, in order to introduce a Linear Dynamical System (LDS) approach to BCI.…”

“…However, we obtain higher accuracy for both subjects, as already indicated. We choose to compare our results with those in [39] as they provide results using similar metrics on the same datasets. Table 2, as well as with other approaches based on BS [35], RQNN [38] and DDFBS [36].…”