Gated adaptive graph causal attention decision capsule transformers network with Meerkat optimization algorithm for EEG- and EMG-guided myoelectric control in upper limb rehabilitation

EEG- and EMG-based Myoelectric Control (MC) in the rehabilitation of the upper limb employs muscle and brain signals to drive assistive device, facilitating motor recovery through adaptive, patient-centered therapy and real-time feedback. Conventional methods tend to be inaccurate, non-adaptive, and lacking in real-time responsiveness because they are limited by signal processing and feature extraction capabilities. To address these, this paper proposes a Gated Adaptive Graph Causal Attention Decision Capsule Transformers Network with Meerkat Optimization Algorithm (GAGCADCTNets + MOA) for EEG- and EMG-guided myoelectric control in upper limb rehabilitation. The input EEG signals are drawn from the BCI Competition IV 2a dataset. EMG data are collected from the NinaPro DB2 dataset and are first preprocessed with the Trainable Self-Guided Filter (TSGF) technique. Then, feature extraction is performed with the Quadratic Phase Quaternion Domain Fourier Transform (QPQDFT). For motor control signal prediction, GAGCADCTNets is used, and then the Meerkat Optimization Algorithm (MOA) is used for optimization. The BCI Competition IV 2a and NinaPro DB2 datasets are used to compare the performance of the proposed GAGCADCTNets + MOA model, which has an excellent accuracy rate of 99.9% along with a recall of 99.8%. The model designed is developed using the Python programming language. The results of the suggested model prove outstanding performance in improving motor signal prediction, outperforming in accuracy, robustness, flexibility, and real-time responsiveness, thereby greatly enhancing upper limb rehabilitation therapy outcomes.

Hali tarjima qilinmagan