Sparse representation-based motor imagery EEG classification towards asynchronous BCI systems

Most of the existing motor imagery (MI)-based brain-computer interface (BCI) systems operate in synchronous to the system-generated time slots. But in real-world applications, users want to control the interface asynchronously at their own convenience. The main challenge in such asynchronous BCIs lies in the detection of relax period. In this study, sparse representation-based classification (SRC) scheme is proposed for asynchronous BCI systems. The dictionary needed for the SRC scheme is learned from the extracted EEG features using the K-SVD algorithm. The proposed framework is evaluated on two benchmark datasets from BCI competitions III and IV. The results showed the SRC's detection ability to relax states and to MI states, which is better than the detection ability of the well-known linear discriminant analysis classification method. The betterment of the proposed scheme is also shown in terms of accuracy while classifying the left-hand MI, right-hand MI, and the relaxed state.