Hand Gesture Recognition Using ECANet via Surface Electromyography Signals

In recent years, with the continuous advancement of intelligent interaction technology, surface electromyography (sEMG)-based gesture recognition has rapidly emerged as a key approach to enabling natural and efficient human-computer interaction. This technology demonstrates significant potential in fields such as medical rehabilitation, assistive device control, and virtual reality. This paper proposes HGRUA-Net, a gesture recognition method incorporating an efficient channel attention mechanism (ECANet). Built upon two convolutional layers and a max-pooling layer, this approach leverages ECANet to automatically focus on the most discriminative features within the signal while effectively suppressing redundant information, thereby enhancing recognition accuracy and robustness. Experimental results show that on the Ninapro DB2, DB5, and Myo datasets, the recognition accuracy of HGRUA-Net without the attention mechanism is $77.09 \%, 64.71 \%$, and $96.68 \%$, respectively. With the integration of ECANet, the accuracy improves by $4.70 \%$, $1.3 \%$, and $0.25 \%$, respectively.