Recognition And Control Of Compound Motion Pattern Of Myoelectric Prosthesis Based On Embedded Linux

In recent years,the demand for prostheses on the market has increased significantly due to traffic accidents and own diseases.For patients with upper limb disability,on the one hand,the disability of the upper limb makes the patient unable to take care of themselves,on the other hand,it also causes severe psychological pressure on the patient.With the development of modern technology and the increasing social attention to patients with residual limbs,various types of myoelectric prostheses have appeared on the market,alleviating the life needs of patients to a certain extent.However,the current electromyographic prostheses still have a series of significant problems.For example,most of the existing prostheses recognize a single movement,and cannot recognize a compound movement.Realizing a specific behavior requires several parts of a single movement to be executed one by one in chronological order.Conflicts with human behavior and habits;at the same time,the existing prostheses need to be trained by professionals before they are used to generate a usable EMG pattern recognition model.This process needs to be completed on the PC side and needs to be regular and frequent It is inconvenient to carry out,and it also increases the economic burden for the patient.The main research content of this paper is to construct the EMG signal pattern recognition model,complete the software and hardware system design and the transplantation of the recognition model on the system,and finally realize the design of the EMG prosthetic compound action pattern recognition and control system based on embedded Linux.According to the characteristics of the EMG signal,a signal conditioning circuit is designed to collect EMG on the flexor carpi ulnaris,superficial digital flexor,palm longus and extensor digitorum.The collected EMG signals are preprocessed,and the characteristic values in the time domain,frequency domain and time-frequency domain are extracted respectively.In order to ensure the accuracy of classification and increase the speed of the system,the XGBoost algorithm is used to extract feature values and recognize patterns of EMG signals.For different actions,extract the time-domain variance,time-domain mean square error,average power frequency,median frequency,maximum wavelet packet coefficient energy,and maximum wavelet packet coefficient variance with the largest correlation coefficients.Features with low correlation are not considered.To achieve the purpose of dimensionality reduction for subsequent algorithms.At the same time,import actual sample data and compare the XGBoost training model with the PCA-LSTM algorithm to verify the superiority of the model.The results show that the feature extraction and pattern recognition model based on XGBoost has faster pattern recognition speed and higher recognition rate for compound actions.Build embedded system,design QT interface and transplant.The interface mainly includes 8 action buttons(4 single actions and 4 compound actions).Each button is pressed to play the corresponding video and guide the patient to perform the corresponding action.When the button is pressed,the EMG signal collection starts.After preprocessing and feature extraction,the collected EMG signals are input into XGBoost algorithm to generate a patient-specific pattern recognition model.When the patient uses the EMG prosthesis,the collected EMG signal is classified by the model,and the corresponding action type is obtained,and the motor is driven to complete the corresponding action.In this paper,through the design of the software and hardware system and the transplantation of XGBoost on the embedded Linux system,the design of the electromyography prosthetic compound action pattern recognition and control system is completed.