An sEMG-based Hand Motion Recognition Method for Stroke Patients with Feature Engineering and Cascade Forest

A hand motion intention recognition method with a high recognition rate and suitable for patients of different rehabilitation levels is proposed to solve the problem of low recognition rate of motion intention of stroke patients based on surface electromyography (sEMG) signal. Firstly, the sEMG data from 30 patients of different rehabilitation levels are used to carry out feature extraction based on tsfresh library and feature selection based on Feature-Selector library, and the most suitable sliding window parameters and the features suitable for the patient motion recognition task are determined. Then, some action recognition experiments are conducted with the proposed method, which is compared with other methods, including random forest, convolutional neural network, etc. The experimental results show that the average recognition accuracy of the proposed method for 9 commonly used hand rehabilitation actions is 97.94%. Finally, a hand rehabilitation system is developed based on the proposed method, its real-time performance is analyzed based on online recognition experiments, and a patient tracking experiment is designed to verify the effectiveness of the system for patient hand rehabilitation.