Feature extraction of Motion Imagination EEG based on S transform and CSP

摘要

Although the CSP algorithm has been extensively studied, little has been known about its relatively low accuracy and efficiency. Based on the evident advantage of S transform in feature extraction, including more obvious characteristic of time, frequency and phase, it is applied to preprocess the data for a better classification. The experiment result on the motion imagery indicates that the combined algorithm achieve an accuracy of 92.8%, far higher than the accuracy of applying CSP algorithm only. What is more, the time cost could be diminished to 0.85s after sampling, 31% less than the original method, and could still achieve the accuracy of 89.8%. In conclusion, the combination of CSP and S transform algorithm is a feasible and applicable one with data and technical support.