Research On Deep Neural Network-based Gesture Command Recognition Using EMG Signal

With the development of artificial intelligence technology,gestures are becoming more and more important as the information source in the process of human-computer interaction systems.The current gesture recognition technology is mainly divided into two types according to the different input devices: vision-based gesture recognition and wearable sensor-based gesture recognition.The latter is less affected by environmental factors and has high stability,making it important for gesture recognition.the way.Wearable sensors mainly analyze the action patterns of gestures by collecting electromyographic signals and inertial signals during hand movements.In this paper,surface electromyography(sEMG)and inertial signals are used as signal sources,and a deep neural network-based gesture recognition model is designed,and the effectiveness of the recognition algorithm is verified through 10 gestures.The main work of this paper is as follows:1.Analyze the characteristics of sEMG,combine the structure and local anatomy of the wearable collection device,determine the action mode and collection experiment,complete the collection of sEMG and inertial signals of 10 actions,and carry out the detection of the active segment.2.Aiming at the problems of low accuracy and complex models of existing gesture recognition algorithms,starting from deep neural networks,three network models of one-dimensional Convolutional Neural Networks(CNN),Long Short-Term Memory Networks(LSTM),and Gated Recurrent Units Network(GRU)are built and experiments are carried out.The recognition rates are 81.39%,85.22%,and 84.97%,respectively.3.Combining the advantages of CNN and GRU networks,build a CNN-GRU network model,and design orthogonal experiments to parameterize the learning rate,batch size,and number of hidden layer neurons of the CNN-GRU network.After optimization,it is finally determined that when the three parameters are 0.0005,30,and 125,the model effect is better,the classification accuracy rate reaches 93.42%,and the recognition effect is better than that of a single network model.4.An online gesture recognition system is designed.The sEMG and inertial signals are classified online through the trained model,and control instructions are sent from the host computer to the electric wheelchair via Bluetooth,and the wheelchair is controlled to complete 10 actions.The control accuracy rate reaches99.2%.The research results of this paper extend the current simulation research to the development of actual application systems,and provide an effective solution for the wide application of gestures in human-computer interaction systems.