An EEG Feature Extraction Algorithm Based on Adaptive Time-frequency Motor Imagery

Abstract

To solve the deficiency of traditional common spatial pattern feature extraction algorithm due to the changes of neurophysiology in subjects with the variation of time and frequency domain, a self-adaptive time-frequency domain common spatial pattern feature extraction algorithm is presented. The algorithm finds out the personalized time and frequency domain of the subject automatically by a data-driven method. The results demonstrated that the algorithm improves the performance of the traditional common spatial pattern. In addition, the regularization term of the algorithm is designed. The experiment results on two public competition data sets show that the classification accuracies of the proposed adaptive feature extraction method is higher than those of the original public model (10.9% and 20.33%).