Multiband Based Joint Sparse Representation for Motor Imagery Classification

The feature extraction method based on multiband has been successfully applied to the classification task of motor imagery (MI). However, it is still a challenge to eliminate redundant information and extract significant common features between multiband. The joint sparse model (JSM) performs joint decoding through joint sparsity among multi-channel signals and has achieved great success in computer vision and pattern recognition. In this work, we employ the Pearson correlation-based channel selection method to establish high-quality spatial distribution. Moreover, a novel multiband based joint sparse representation (MJSR) is proposed to fuse CSP features of multiband and obtain joint coefficient features. The SVM is then applied to classify MI tasks. The experimental results on two BCI competition datasets indicate that the proposed method can effectively improve the classification performance.