A Study on the Classification Method of Motor Imagery EEG Signals Based on Band Filter Bank Common Space Pattern and Continuous Wavelet Transforms

EEG signal feature extraction link will affect the accuracy of MI, this paper is optimized for feature extraction to improve the classification accuracy. A feature extraction model (Band Filter Bank Common Spatial Pattern-Continuous wavelet transform, CCWT) is proposed to address the problem that the airspace features extracted by Common Spatial Pattern (CSP) are relatively homogeneous. The model is based on the band filter bank co-space pattern fusion Continuous wavelet transform (CWT) algorithm. The original data are first divided into multiple frequency bands, and the spatial domain features are extracted by CSP, and then the features are processed by continuous wavelet transform, so that the features contain time-frequency domain information, and the final feature matrix is obtained. The study validates the proposed method CCWT on the public dataset BCI Competition IV 2a dataset with an average accuracy of 86.713% and an average Kappa of 0.8231.