The EMG Pattern Recognition System Based On Generalized Dynamic Fuzzy Neural Network

In modern rehabilitation medicine, there is an increasing demand of medical system with the function of intelligent rehabilitation, and it is one of the indispensable procedure to create the system by recognizing the sports consciousness of people accurately by the electromyographic (EMG).The content of the thesis includes the research of characteristics of EMG, how to record EMG, how to extract the features, how to recognize the action. According to the compare of several algorithms of feature extraction, we combine it with the pattern recognition algorithms to establish a stable experimental system, and certificate the effect of this system using actual EMG signals, finally recognize the multi-action modes of hand.At the beginning, the paper introduces the current progress in the research of EMG pattern recognition and its related fields. Then we take a systematic approach to several algorithms of EMG feature extraction and their significance is given. After that, the paper summarizes the structural features and mechanism of the fuzzy system and neural network, introducing the weight operations, membership function selection, basic structure of fuzzy neural network (FNN) which incorporating their merits, and a network structure of a FNN with dynamic adjustment feature(DFNN) is emphasized. Then the structure of the EMG collecting device including the hardware and software is introduced. Afterwards, the paper describes the standards and implementation procedure of the algorithms of pattern recognition based on generalized dynamic fuzzy neural network (GDFNN) in detail. At last, this paper provides an analysis and comparison in seven EMG feature extraction algorithms, and summarizes their results of pattern recognition. The experiments indicate that the accuracy could be 97% by using algorithms of GDFNN to recognize seven type of actions and it has a wide applicability in several approaches of feature extraction.By the development of this experimental platform, we proposed implementing approaches of hand's multi-action pattern recognition based on EMG signals. In the future, we will search a more suitable method of EMG collecting, studying the new algorithms of feature extraction. At the same time we will adjust the algorithms of pattern recognition to improve the real-time performance of the system, laying a solid foundation for motion control of rehabilitation robot.