Research On Manipulation Mode Recognition And Grab Force Prediction Technology Based On Surface EMG Signal

The surface EMG signal is a complex superposition of the action potential generated by a plurality of moving units raised during muscle excitation. Currently, intelligent prosthetics based on sEMG has become a hot topic. Unfortunately, most studies have tended to identify manpower operation mode, but rarely prosthetic hand or the closing force manpower crawl Objects should be applied to predict the estimate, making it difficult to come up to complete the task more accurately crawl intelligent prosthetic hand. This article aims to explore the method of sEMG-based hand gesture recognition and crawl gripping force prediction. The main contents of this paper are:(1) Proposed 6th order Butterworth band-pass filter and fast independent component analysis(Fast ICA) combining method preprocesses sEMG; wavelet packet combined sample entropy feature extraction methods obtains a standard sample entropy sEMG(SSE) feature; using SVM classifier model based on two-channel and four-channel sEMG crawling movement experimentize on pattern recognition.(2) Using AGA-ε-SVR prediction model respectively based on predetermined and random crawl mode, with the standard sample entropy characters, experimentizes on prediction of gripping force. Prediction accuracy is superior to the traditional which uses of sEMG’s amplitude or MAV features to grip force’s prediction.(3) MATLAB-GUI experimental platform is designed to detailedly analysis accuracy of pattern recognition and crawl gripping force prediction, and verifies the feasibility of the program:Verification experiment of based sEMG hand crawling motion recognition platform, the result shows correct recognition rate of the two channels based on four kinds of operation modes crawl is greater than 92%. Further cross-validation test results shows the lowest recognition rate is 90% verifies the good performance of the classifier. Give each fetch operation mode based on the experimental results shows recognition rate based four channels is greater than 96%, indicating that an appropriate increase in the number of electrodes resulting recognition rate increased. In the crawl mode based on arbitrary power prediction results show that the amplitude of the second channel sEMG gripping force can reflect the change in size; regression accuracy AGA-ε-SVR forecasting model is superior to BP algorithm prediction model; based on four-channel experimental results show that the prediction of force prediction accuracy is better than force prediction based on the two channels. Under the provisions on force fetch mode prediction results show that its four-channel power prediction accuracy of better than random prediction force crawl mode; based on MAV features four-channel sEMG were force prediction accuracy is superior to direct force conducted for prediction of sEMG amplitude, while the use of SSE feature sEMG were force prediction accuracy is better than the forecast carried out by the force of the MAV features sEMG. Experimental results show that the arm of sEMG signal not only able to identify the type of shot mode movements, but also to predict the size of the gripping force, the study will help develop accurate crawling tasks to complete myoelectric prosthetic hand.