Shrinkage estimator based-DGLRSQ for motor imagery classification

Motor imagery classification is a crucial task in the field of brain-computer interfaces (BCI)which can benefit individuals with disabilities in their movements of their limbs. However, accurately classifying such signals poses a significant challenge owing to the complexity of the signal dynamics. This paper proposes a approach for motor imagery classification utilizing shrinkage estimator-based dynamic generalized learning Riemannian space quantization (DGLRSQ). Our proposed approach aims to provide a more accurate estimated covariance matrix by utilizing a shrinkage estimator, which avoids overfitting through data regularization. Additionally, the impact of the sliding window size on the classification performance is investigated, where shorter shifting size delivers better classification performance but with lower computational efficiency. Experimental results demonstrate the efficacy of our method and its superior performance on the publicly available BCIIV 2a motor imagery EEG dataset.