Study On Wrist Movement Recognition Algorithm For Stroke Rehabilitation Training

In today’s society,stroke causes many people to be disabled for a long time.Stroke patients have wrist function loss after the onset of the disease,unable to achieve some basic wrist movements such as wrist flexion,adduction and abduction,which seriously affects the daily life of patients.Using rehabilitation robot for wrist rehabilitation training is one of the important means to promote the rehabilitation of patients.The rehabilitation robot can monitor the muscle electrical signals of the upper arm and wrist of the patients in real time,send out movement instructions through the control system,promote the functional reorganization of the patients’ nervous system,effectively alleviate the atrophy of muscles and joints,and help the patients recover their body functions.Among them,the motion command issued by the control system is determined based on the correlation model of muscle electrical signal and wrist motion,and the accuracy of the model determines the effect of robot rehabilitation system auxiliary training.On the other hand,different patients or patients in different stages of rehabilitation can bear the range of motion is also very different.Therefore,it is necessary and meaningful to build a highprecision muscle electrical signal and wrist motion model.This thesis aims at the problem that different patients or patients in different rehabilitation stages need to adjust the auxiliary training scheme in real time in the current wrist rehabilitation robot training,combined with the idea of stage rehabilitation training(that is,patients with poor endurance first carry out rehabilitation training with small range of activities in the early stage,and then change to large range of exercise training in the later stage),By establishing a network model structure between the patient’s EMG signal and the wrist movement position,the rehabilitation robot can carry out staged auxiliary movement according to the patient’s own condition in real time,and can also be used for rehabilitation training for different patients.Aiming at the problem that the recognition accuracy of one-dimensional muscle electrical signal is relatively low in the past methods,this thesis proposes a classification and recognition method based on the combination of symmetrical point pattern(SDP)and deep convolution neural network,and constructs a high-precision classification and recognition model of wrist movement and range of motion based on muscle electrical signal.Firstly,nine kinds of muscle electrical signals related to wrist movements in ninapro database are transformed by SDP method.The transformed samples are input into vgg-16 neural network model,and the classification accuracy is 95%.Compared with the previous one-dimensional data,the better results are improved by about 3%.In order to verify the recognition effect of the proposed model,an experimental platform for rehabilitation training signal detection is built,which includes wrist rehabilitation robot,s EMG signal sensor and inertial measurement unit.Inertial measurement unit(IMU)is used to acquire the attitude information of wrist motion in real time.The experimental platform was used to collect eight kinds of s EMG signals and the corresponding attitude position information,and the database of s EMG signals and their corresponding attitude position was established.The surface EMG signals were transformed by SDP,and 7264 of them were used as training samples and 800 as test samples.The vgg-16 and resnet-50 deep neural network models were established.The range of motion of each of the four wrist movements is divided into two categories to represent the difference of patients’ tolerance.The test shows that vgg-16 model can achieve 94.8% recognition accuracy and resnet-50 network model can achieve 99%recognition accuracy.