Feature-Fused Deep Learning Approach for Hand Gesture Recognition in Intelligent Myoelectric Hand

Accurate and efficient hand gesture recognition is a cornerstone for advancing the functionality of myoelectric prostheses, enabling intuitive and responsive human-machine interaction. In this study, we present a novel hybrid deep learning architecture that synergistically integrates convolutional neural networks (CNNs) with Transformer models to achieve unprecedented accuracy in electromyogram (EMG)-based hand gesture classification. By leveraging advanced feature engineering, we extract and combine time-domain (TD), frequency-domain (FD), and time-frequency (TF) features to provide a comprehensive representation of muscle activity patterns. This approach addresses critical challenges in existing methods, such as poor generalization and limited robustness to subject variability. The proposed system achieves a state-of-the-art (SOTA) recognition accuracy of 98.5% and an F1-score of 0.98 across 12 distinct hand gestures, surpassing existing methodologies. Additionally, we validate the practicality of our system by implementing it in a real-time control platform for a myoelectric hand, demonstrating its capability to perform precise, real-world motor tasks. Our findings represent a significant step forward in neural systems and rehabilitation engineering, offering a transformative solution for enhancing the usability and functionality of intelligent myoelectric prostheses.