Gesture Recognition Based On Pattern EMG

Surface Electromyographic(SEMG)is an electrical signal produced by muscle contraction and is an external representation of human tissue activity.Therefore,it is an excellent natural interface for human-machine interaction,and has been widely applied to the field of the myoelectric hand.The basic principle of the myoelectric hand is that pattern recognition technology is be uesd to classify the SEMG collected from upper limb,and the mechanical hand is driven to complete the corresponding gestures according to the results.Although the current myoelectric hand has realized the recognition of a variety of gestures,there are still various shortcomings:(1)the degree of freedom of gestures studied in the past is generally small so that the current myoelectric hand can not meet the basic daily needs;(2)the motion switching is not smooth enough which resulting in a bad experience.These shortcomings seriously affect the application and development of the myoelectric hand.Therefore,this paper studies and designs a recognitional system based on EMG that can recognize 8 common gestures and explores the smooth switching.The main work of this paper is as follows:(1)Review the research status of the myoelectric hand at home and abroad,and focus on the research status of myoelectric control system based on pattern recognition technology.(2)Understand the mechanism of myoelectric signal generation and the muscle function of the human forearm.Select the muscle groups associated with the 8 gestures as the data acquisition channel,and use the NORAXON acquisition system to complete the collection of SEMG.(3)Denoise SEMG which includes noise to improve the signal-to-noise ratio of the signal.In the filtering process: the Butterworth bandpass filter is used to eliminate the baseline noise and wavelet coefficient threshold denoising method based on the Gaussian mixture model is used to eliminate the background noise.(4)Use time domain and wavelet domain analysis methods to extract features from SEMG.In the feature selection process,the random forest method is used to calculate the relative importance of each feature,and the feature is selected for its relative importance.(5)In the research of pattern recognition classifier,two classification algorithms which are support vector machine and BP neural network are put forward and optimized.(6)Study classification effect of two types of SEMG which are collected from the process of continuous motion and the process of motion switching.By smoothing the "judgment flow",the recognition accuracy of two process can be effectively improved.The real-time recognition system designed in this paper provides a feasible reference solution for the practical application of the myoelectric hand.