Exoskeleton Human-Computer Interaction Based on Deep Learning and sEMG Imaging
Year: 2024
Type: conference-paper
Abstract: Methods: Starting with the rapid recognition of human gait events, limb joint movement angles, and active limb joint moments. Gait data were collected from six healthy subjects performing extended squat, in-situ trot and jump movements on a 3D force platform, with each movement performed 20 times. Calculations of human hip, knee and ankle joint moments were output by the exoskeleton and MAS respectively, and the collected gait data was used for training and validation. The experimental data was quantified by Root-mean-square error (RMSE), and Correlation coefficient (CC) for comparison. Results: A technique for sEMG signal analysis based on deep learning and integrated learning techniques is proposed. Using the raw sEMG signal for direct classification simplifies the feature extraction process and achieves a high accuracy classification with an area under the hit curve (AUC) value of 0.9554. A fast-training sEMG signal feature classification method based on integrated learning techniques is proposed, using motor-related cortical potentials as well as neural oscillations of different frequencies as features to obtain a classification algorithm that exceeds the accuracy of any of the base classifiers in the same time-training time. The results show that joint motion recognition and prediction can be effectively performed using this model, validating the feasibility of this system as an exoskeleton human-machine interface. Conclusion: Exoskeletal robots can provide safe and reliable rehabilitation for patients with neurological disabilities such as strokes. The advantages presented by the optimisation of exoskeletal robots make it possible to replace traditional rehabilitation therapy.
Authors: Wanni Xu, Minghua Guan, You-Lei Fu, Weifang Xie
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4406356444