Research On Hand Motion Recognition Based On SEMG Signal

The surface EMG signal(sEMG)is a weak bioelectrical signal generated on the surface of the skin during muscle contraction.Due to the characteristics of convenient collection,non-invasiveness and no damage to the limb,it has been extensively studied in recent years.Interactive devices controlled by sEMG have potential application research value.Research on human hand motion recognition based on sEMG has become an important research hotspot of bionic prosthesis.Although some progress has been made in current sEMG-based motion recognition,there are still some difficulties on the analysis and use of sEMG envelope signal,the complexity and accuracy of active segment detection algorithms,the optimization of feature selection and processing,and individual differences,which causes a lower recognition rate and lesser accurately identified motions.In view of these problems,this paper aims to develop sEMG-controlled prosthesis,and adopts the method of combining experiment and theory to conduct the following research on human hand motion recognition based on sEMG signals:1)The characteristics of the two types of sEMG signals,including the sEMG raw signal and the sEMG envelope signal,are investigated.The former is obtained directly from the skin surface through the silver chloride electrode paste,and only through simple amplification without further processing by the signal processing circuit.It is an AC signal with rich time domain and frequency domain information.The latter is the DC signal obtained by rectifying,integrating and amplifying for sEMG raw signal using a signal processing circuit,which reflects the envelope track of sEMG original signal and carries abundant information in time domain.2)The active segment detection and preprocessing methods are studied.In the process of active segment detection,this paper proposes and compares improved algorithms based on short-term energy and sliding absolute value average,and analyzes their advantages and disadvantages.About data preprocessing,this paper adopts eight-order butterworth bandpass filter for the sEMG raw signal,and studys three preprocessing methods for the sEMG envelope signal,including sliding median filtering,digital filters and wavelet transforms.Finally,the paper selects the smoothing and denoising preprocessing technology based on six-layer decomposition and reconstruction of wavelet.3)The sEMG feature extraction and processing methods are studied.These research contents include feature extraction,feature preprocessing and feature selection optimization.According to the nature of sEMG raw signal,52 features are extracted from the time domain,frequency domain,time-frequency domain and parameter model.According to the characteristics of the sEMG envelope signal,24 features are extracted from the time domain.Different features have different magnitude.In the preprocessing of features,this paper carries out the maximum and minimum scaling processing for all features to eliminate magnitude.Due to individual differences,the optimal feature set for different subjects is not the same,and this paper uses four univariate feature selection algorithms,including Pearson,F-test,Chi-square test and Relief-F,in order to select the optimize feature subset.4)The use of sEMG and the matching between supervised learning and univariate feature selection are studied.This paper uses three supervised learning algorithms,including K-Nearest Neighbor,Multilayer Feed-Forward Neural Network and Support Vector Machine.In the experiment,it is found that the recognition rate is relative to matching of the classifier and the univariate feature selection algorithm.At the same time,the combination of the sEMG raw signal and the sEMG envelope signal performs best.After a series of analysis and comparison,this paper selects the matching of F-test and Support Vector Machine,and the method of combination of the two signals.In this way,this article uses only two sensors and achieves a 95% recognition rate for nine hand motions.5)The online real-time sEMG-controlled bionic hand experiment platform is designed.The conclusion obtained in the above chapters is realized on the hardware platform,and a real time experimental platform is developed for the bionic hand to imitate experimenters.The experimental platform includes a data acquisition terminal,a waveform display and data processing terminal,and a bionic hand control terminal.They communicate with each other through the serial port or the NRF24L01 wireless communication module,which can online identify and simulate the experimenter's actions in real time.For details,please refer to the appendix.