Development of a Deep Learning-Enhanced Lower-Limb Exoskeleton Using Electromyography Data for Post-Neurovascular Rehabilitation
Abstract
This paper presents the development of a deep learning-enhanced lower-limb exoskeleton designed to facilitate post-neurovascular rehabilitation using electromyography (EMG) data. The system leverages real-time EMG signals to classify user movement intentions, such as leg up, leg down, and leg move, and translate them into corresponding motor actions. A pretrained deep learning model was utilized to accurately classify muscle activity, while the Short-Time Fourier Transform (STFT) was employed for feature extraction from EMG signals, enhancing the distinction between different types of movements. The results demonstrate the system’s high classification accuracy and real-time adaptability, as evidenced by the confusion matrix and performance metrics. The exoskeleton's motor control system successfully provided appropriate mechanical assistance based on the classified movements, offering a personalized rehabilitation experience. Despite these positive outcomes, further improvements in generalization, such as increasing the range of movements and applying the system to a larger, more diverse dataset, are necessary. Future work will also explore the integration of additional sensor modalities to improve classification performance. Overall, the proposed exoskeleton demonstrates significant potential to enhance rehabilitation outcomes by actively engaging the user in task-specific training, contributing to improved neurovascular recovery.