An sEMG Gesture Recognition Algorithm Based on Improved MobileNet

Surface electromyography (sEMG) signals are generated by the potential changes of muscle fibers. Gesture recognition based on sEMG is a research hotspot in action recognition. And it has been widely used in many fields such as rehabilitation medicine and human-computer interaction. Convolutional neural networks, a typical class of deep learning neural networks, are often applied to sEMG gesture recognition currently. However, these methods have problems of too many parameters, large computation, and long running time. To these problems, MobileNet, a lightweight network model with fewer parameters, is adopted in this paper, and an sEMG gesture recognition algorithm based on improved MobileNet is proposed. The improved MobileNet uses MobileNet as the main architecture, and introduces a genetic algorithm to optimize the parameters in the MobileNet. Simulation results show that, the proposed recognition algorithm reduces the parameter amount and the computation burden without loss of accuracy.