Motion Recognition of Bionic Manipulator Based on Surface Muscle Electrical Signals

This paper studies a bionic gesture recognition system based on surface electromyography (sEMG). The system was designed and realized based on STM32F4. The sEMG signals of the operator's upper palmaris longus muscle, extensor digitorum muscle and flexor digitorum superficial muscle were collected by means of electrode patch. The machine learning method was used to improve the quality of signal acquisition, optimize motion recognition, improve motion recognition accuracy and control the manipulator to make corresponding actions. In this paper, a set of gesture recognition data set is constructed, which contains 90,000 data of 24 kinds of gesture actions. Through comparative analysis of BP, MPL, LeNet and DenseNet, it is shown that the system can obtain better recognition accuracy by using the MPL model and the LeNet model. In addition, a control experiment was conducted in this paper. The experimental results show that the recognition accuracy of the system can be significantly improved when the gesture data of the experimenter is added to the data set for training.