Exoskeleton Human-Computer Interaction Based on Deep Learning and sEMG Imaging

This research develops an optimized exoskeleton design for human-robot interaction, using surface electromyography (sEMG) to detect patient movement intentions and motor functions. The study involved rapid identification of human gait events, joint angles, and movements, using data from six individuals performing various movements on a 3D force platform. The exoskeleton and motion analysis system (MAS) processed this data, which was utilized for model training and validation. The evaluation was conducted using Root-mean-square error (RMSE) and Correlation Coefficient (CC). We introduced a novel sEMG analysis method leveraging deep and integrated learning to directly classify sEMG signals, achieving high classification accuracy (AUC of 0.9554). A fast-training classification technique utilizing motor-related cortical potentials and varied neural oscillations also exceeded the accuracy of existing classifiers, effectively supporting joint motion recognition and prediction. The optimized exoskeleton provides a viable and safe rehabilitation alternative for patients with neurological disabilities, offering a potential replacement for conventional therapies.