Subjects self-select forearm gestures identification based on post-processing and integrating sEMG sensors and AS

• Published in: Biomedical signal processing and control Vol. 85; p. 105023
• Main Authors: Zhang, Lei, Bian, Huarui, Wang, Jie, Wang, Along, Zhang, Kangkun, Pang, Ming, Liu, Hui
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.08.2023
• Subjects: Forearm gestures; Machine learning; Post-processing; sEMG; Forearm gestures; sEMG; Post-processing; Machine learning
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2023.105023

•The use of sEMG signals to recognize movement intention has important applications in artificial hand control and rehabilitation training for individuals with hemiplegia or muscle weakness. However, sEMG signals are weak and susceptible to external interference, which can limit their accuracy in recognizing specific movements. To address this challenge, this study proposed a classification method in which participants could voluntarily choose their forearm movements.•The study used two sEMG sensors and a 9-axis attitude sensor attached to the wrist to record the selected movements. Nine participants were included in the study, each selecting five movements. The results showed that the K-Nearest Neighbor (KNN) algorithm performed best, with a recognition accuracy of 96.2%.•Overall, this study demonstrates the potential of using a participant-selected movement classification method to improve the accuracy of sEMG-based recognition of movement intention. By allowing participants to select their movements, the approach could increase the adaptability of sEMG-based purpose recognition strategies and play an important role in advancing artificial hand control and rehabilitation training. The use of surface electromyography (sEMG) to determine movement intention has a lot of promise regarding artificial hand control and hemiplegia rehabilitation. Nevertheless, since sEMG is fragile and vulnerable to outside intervention, the current study focuses on identifying particular postures. When the subjects are swapped out, the recognition accuracy plummets. This study proposed a method in which the participant could select their forearm gestures on a voluntary basis. Nine subjects' selected movement datawere recorded using two sEMG and nine-axis attitude sensors. This paper used post-processing to optimize the features, which were then identified using K-Nearest Neighbor algorithms to increase recognition accuracy. The combination of the two, as well as post-processing, could improve the recognition effect with 96.2 ± 6.9% accuracy, which was a statistically significant difference from the other way. The proposed model couldbe employed as a user-independent movement classification, with these subjects being able to choose forearm movements independently. The proposed approach can potentially increase the adaptability of sEMG-based purpose recognition strategies and play a key role in popularizing manipulators or prosthetic control and recovery training.