Gesture Recognition Based On SEMG Signal

Surface electromyography(sEMG)is a weak bioelectric signal generated during muscle activity.It contains a wealth of information.It is widely used in the fields of sign language recognition,artificial hand control and human-computer interaction.The precise movement of muscles is related,so it can be used to identify the movement intention and state of the limbs.Although some progress has been made in gesture recognition based on surface EMG signals,due to the non-linear and non-stationary characteristics of surface EMG signals,the noise reduction,active segment detection,feature extraction and feature selection of surface EMG signals are all It has been optimized.There are still some problems.There are not many gestures that can be recognized,and the recognition accuracy is not ideal.This paper uses a combination of theory and experiment to process the original surface EMG signal.The details are as follows:(1)At present,there are few types of gestures in the research of surface EMG signal pattern recognition,and the effect of pattern recognition of tiny finger gestures is not ideal.In this paper,we have designed 12 commonly used gestures based on the actual situation,and by comparing the role of each arm muscle group in gestures,it is determined that the electrodes are placed on the palmar longus,flexor carpi radialis,flexor carpi ulnaris and finger.The position of the extensor muscle belly.(2)As the collected surface EMG signal usually contains a lot of noise,this article compares the noise reduction effects of different wavelet thresholds and threshold functions,and finally selects the sym8 wavelet function to perform 6-layer wavelet decomposition on the surface EMG signal,and uses heuristic threshold and soft threshold Function for noise reduction.In order to effectively extract activity segment data,this paper compares three types of activity segment data extraction algorithms.The experimental results show that the dual-parameter multi-threshold algorithm can effectively extract active segment data.(3)Previous studies mostly selected several constant features for gesture recognition,ignoring the influence of redundant features on gesture recognition.This paper uses time domain,frequency domain,time-frequency domain multi-domain fusion combined with minimum redundancy and maximum correlation,filtering algorithm based on fast correlation,Relief-F and Pearson correlation coefficients for feature selection.Combining support vector machines and linear discriminant analysis classifiers with 4 feature selection algorithms to obtain different recognition models,it is found that different recognition models get different results,and the gesture recognition model based on multi-domain fusion and feature selection can be used in fewer features In this case,a higher recognition accuracy rate is achieved.(4)The surface EMG signal is a non-linear,non-stationary random signal.Extracting good features is a prerequisite for obtaining a higher recognition accuracy.This paper proposes an improved wavelet packet feature extraction method.First,empirical mode decomposition is used to smooth the signal,and then the basic mode components with higher correlation with the original signal are selected for wavelet packet transformation,and the wavelet packet coefficient features are extracted,and the extracted features The dimensions are different,and the maximum minimum method is used to normalize the features.Experimental results show that the improved feature extraction method improves the recognition accuracy by 8.67%.