EMG-Based hand gesture recognition using temporal convolution networks

Abstract

Electromyography (EMG)-based hand gesture recognition has become a pivotal technology in the development of intuitive and responsive human-computer interaction systems, particularly for prosthetic control. EMG signals, which reflec tmuscl eactivity during motion, offer a non-invasive and effective means of decoding user intent. However, the inherent variability, noise, and temporal complexity of EMG signals present significant challenges to accurate gesture classification. This project explores the use of Temporal Convolutional Networks (TCNs), a deep learning architecture for sequential data, to improve hand gesture recognition from range temporal dependencies efficientl yan dstably .Th epropose dmodel ,traine do nthe benchmark NinaPro dataset, incorporates residual connections and attention mechanisms to enhance learning depth and temporal focus. The results show strong classificatio nperformance ,lo wloss ,an defficie nttraining times. Most gestures were accurately recognized, as reflecte di nth econfusio nmatrix .To counter overfittin gfro minter-subjec tvariability ,w eapplie dextensiv edat aaugmentation, which improved generalization. The fina lmode ldemonstrate sstron gpotentia lfo rreal-time EMG-based gesture recognition, advancing reliable myoelectric prosthetic control systems.