Research On Gesture Classification Algorithm Of Amputee Based On Signal Feature Fusion

With the development of artificial intelligence technology,human-machine interaction technology has developed rapidly in the field of medical rehabilitation.Researchers use surface EMG signals to achieve prostheses that can be controlled by own intentions for amputees,helping amputees to making up the regret of losing limb.As a result of the surface EMG signal is a non-stationary signal,and the amputee's age,the remaining degree of the forearm,the sensitivity of the phantom limb,the time of the amputation are different.No matter the algorithm perspective or the classification success rate pose challenges to the gesture recognition of amputees.In this paper,the published data set DB3 by Ninapro is used to study 17 gesture recognitions of amputees,including static gestures and wrist movements.The purpose is to improve the adaptability,simplify feature extraction algorithms,reduce training samples used and improve the recognition rate.The main research work is as follows:(1)In view of problem that lack of adaptability,complex feature extraction algorithms,and requiring a large number of training samples in the process of surface EMG signal gesture recognition of amputees,this paper uses RMS window smoothing filter to eliminate random fluctuations of the signals and enhance the signal-to-noise ratio of the signal.Then method of intercepting signals by overlapping sliding windows meets the needs of real-time and adaptability,and applying the Gray theory Model(GM)to feature extraction algorithm.Finally,using Support Vector Machine(SVM)classification that supports small sample learning to complete the recognition task.According to single-feature comparison experiment verifies that the feature items of the gray-scale model can effectively represent the gesture intention of the amputees,and fusion of the mean feature overcomes efficiently the problem that the standard deviation of the recognition results is too large.According to the recognition for 17 kinds gestures of 9 amputation subjects,including static gestures and dynamic actions,the average recognition rate is 87.28%,and the time axis error rate is 0.259,it use fewer EMG signal features to obtains a better recognition rate.(2)Aiming at the difference in the sensitivity of amputees' gesture intention recognition,Rely only on surface EMG signal lacks stability and there is still improvement possibility in the recognition rate.This paper proposes a gesture recognition method that fuses the characteristics of surface EMG and acceleration signals for pattern recognition,which improves the stability of the algorithm and the average recognition rate.The recognition result for 17 types of actions of 9 amputee subjects,an average recognition rate of 91.14% and the time axis error rate of 0.2405 were achieved.The analysis of the results showed that the most appropriate size of the sliding window was 250 ms;It was proved the surface EMG signal and the acceleration signal complement each other.When the feature classification of the surface EMG signal is not obvious,fusing the acceleration signal can effectively improve the recognition rate.