Electromyographic hand gesture recognition using convolutional neural network with multi-attention

Abstract

Electromyographic hand gesture recognition can be used to develop myoelectric prosthetic hands for disabled people when the acquired myoelectric signals are accurately classified. Recently, the developed end-to-end deep learning methods are widely used in hand gesture recognition for their automatic extraction of features from raw signals. But not all extracted features are useful and unimportant information can impede gesture recognition. In this study, a deep learning model using multi-attention for electromyographic hand gesture recognition is proposed. The attention mechanism inspired by the biological system enables deep learning models to focus on the important parts of input data while ignoring unimportant information. At first, a convolutional neural network is employed as the feature extractor, enabling the learning of high-level discriminative features from input signals. Then, multi-attention blocks, including channel attention, spatial attention, and temporal attention, are integrated into the convolutional layers, enhancing the extraction of pertinent features by emphasizing critical information through weight recalibration. Finally, the proposed model is evaluated on both the public NinaPro dataset and the public Myo dataset, demonstrating superior performance compared to baseline models in terms of recognition accuracy. The experimental results reveal that the model with attention mechanism achieves an average recognition accuracy of 91.64%, representing a 1.1% improvement over the model without attention mechanism.