Recognition of Four-Class Facial Expression Electroencephalogram Data Using Improved Common Spatial Pattern Algorithm

Aiming to solve such problems as the difficulties in feature extraction, high computational complexity and low signal recognition rate when using traditional common spatial pattern (CSP) algorithm to process electroencephalogram (EEG) signals, an algorithm based on repealed bisection filler bank (RB-FBCSP) and support vector machine (SVM) was proposed. The algorithm is a method used to recognize EEG signals of four types of facial expressions (left smirk, right smirk, frown brown, and raise brow). Firstly, the collected facial expression EEG signals are filtered out by filter bank to select the signals including a wave and 9 wave; secondly, the four types of expressions are regarded as upper facial expressions (frown brown and raise brow) and lower facial expressions (left smirk and right smirk). CSP feature extraction is carried out in two categories, and combined with SVM classifier for classification; finally, the EEG signals of the identified upper and lower facial expressions are repeatedly subjected to CSP feature extraction and SVM two-classification, and the four classifications of auxiliary EEG signals can be realized. The experimental results show that the computational complexity of the proposed recognition method is significantly reduced, the calculation time is shorter, and the average classification accuracy is as high as 89. 614%. Compared to the traditional OVO-CSP, OVR-CSP and wavelet packet transform algorithm combination, the average recognition rates obtained by SVM classification arc improved by 9. 23%, 9. 82% and 8. 045%, respectively. © 2022 Xi'an Jiaotong University. All rights reserved.