Gesture Recognition Algorithm Based on New EMG Representation and Convolutional Neural Network

Gesture recognition based on EMG is a promising method for human-computer interaction. And it is important to improve the accuracy of EMG signal analysis. Current EMG gesture recognition has problems of long training time and low recognition accuracy. To solve these problems, a new EMG representation for gesture recognition algorithm based on convolution neural network is proposed. In this method, the organization of input matrix containing EMG information is improved to reduce the loss of feature extraction. And to enhance the time correlation between adjacent elements in the input matrix, the adjacent data in time domain are put together. Then the LeNet-5 network is applied for signal classification. The classic NinaDB1 data set is adopted in the test. Simulation results show that the method proposed has the advantages of high accuracy and high stability.