Enhancing Hand Movement Recognition: A Hybrid Fuzzy Deep Neural Network Approach with Time-Frequency EMG Representations

This study proposes a novel Hybrid Fuzzy Deep Neural Network (HFDNN) for recognizing complex hand movements using electromyography (EMG) signals. The system integrates fuzzy logic for managing uncertainty with deep neural networks (DNNs) for learning spatial and temporal features of EMG signals. Six hand movements, including lifting heavy and light objects, grasping pen-like and card-like objects, making a fist, and opening the hand, were analyzed. EMG signals were preprocessed using a band-pass filter and normalized, followed by feature extraction through short-time Fourier transform (STFT) to generate time-frequency representations as input to the HFDNN. The proposed model achieved a classification accuracy of $\mathbf{9 6. 8 \%}$, with precision, recall, and F1-scores averaging $95.7 \%, 97.2 \%$, and $96.3 \%$, respectively. Compared to traditional CNN, RNN, and hybrid CNN-RNN models, the HFDNN demonstrated superior performance, validated through statistical significance testing ($\mathbf{p}\lt 0.01$). The inclusion of fuzzy rules enhanced decision-making, particularly for overlapping classes. This work underscores the efficacy of hybrid approaches in capturing the intricacies of hand movement signals, paving the way for advancements in prosthetics control and rehabilitation. Comparative analysis with recent studies highlights the proposed method’s robustness and potential for real-world applications.