Joint feature extraction of bispectral analysis combined with symbol entropy for motor imagery based on BCI

In order to improve the classification performance of EEG based motor imaging BCI, we propose a joint feature extraction method that combines A and B. It mainly analyzes the nonlinear phase coupling of signals and the randomness and predictability of the system, making the separability between the features corresponding to each category more obvious, thereby improving the classification performance. This article focuses on the feature noise extracted by MI-BCI subjects, which is difficult to classify. Inspired by dual frequency analysis and joint feature expression, it aims to explore potential features in multidimensional nonlinear space to maximize data separability, in order to find the information with the maximum separability of features. We applied the proposed method to two publicly available datasets and compared it with the currently widely used and effective feature extraction methods. The experimental results showed that this paper achieved good results on both datasets and made progress compared to the currently effective methods.