Channel-optimized Local Region Riemannian Approach for Motor Imagery Classification

In this paper, we propose a novel motor imagery (MI) classification method that uses channel optimization within local region and Riemannian approach with filter bank to improve classification performance. The proposed method establishes local regions as channel subsets consisting of adjacent channels and selects the optimal channels within each region to generate discriminative local Riemannian features. Unlike traditional global channel selection methods that ignore channel location and discard channels completely, the proposed method preserves local information and allows selective inclusion or exclusion of channels in different local regions. Channels are optimized to minimize confusion area score, a criterion that assesses the misclassification risk of a Riemannian feature based on Riemannian distance distributions, using a backward iterative algorithm. The EEG signals from each channel-optimized local region are decomposed using a filter bank, and a local covariance matrix is computed for each subband. Several matrices are selected based on the confusion area score, mapped to the tangent space, and fed to a classifier. The performance of the proposed method has been evaluated using the publicly available BCI Competition III dataset IVa, and the results show that the optimization of local region channels improves the performance and achieves higher classification accuracy compared to the related existing methods.