Research On Detection And Elimination Of Trunk Compensation Based On Surface Electromyography

Stroke patients often recruit complete trunk muscles and joints to compensate for damaged upper limbs.This abnormal movement pattern is called trunk compensation.The generation of trunk compensation reduces the effectiveness of rehabilitation training and is not conducive to the recovery of damaged upper limbs.Therefore,the automatic detection and elimination of trunk compensation has important clinical application and research value.At present,indirect detection methods based on inertial sensors,cameras,and pressure distribution cushions have been proposed and widely used,and many elimination methods based on force feedback and visual feedback have been formed.However,there are still some shortcomings.First,the existing indirect detection methods cannot characterize muscle movement status,so a direct detection method is urgently needed.Second,the force feedback control strategy was too simple,did not consider the individual differences of stroke patients,and lacked real-time,dynamic adjustment.In view of the above two shortcomings,this paper proposed a trunk compensation detection and elimination method based on surface electromyography(s EMG)signals.A platform for trunk compensation detection and elimination based on s EMG signals was built and verified by experiments.The main research work of this paper is as follows:Firstly,a direct detection method based on s EMG signals was proposed.The s EMG signals of trunk muscles were collected and processed,and a machine learning algorithm was used to realize detection and classification.The results showed that using support vector machine(SVM),the detection performance of healthy participants is better than that of predecessors,with an average accuracy of 95.0%,and an average detection accuracy of 83.1% in stroke participants was achieved.This result proved that the trunk compensation detection method based on s EMG signals was feasible.Secondly,for the problem of low classification performance of stroke participants,this paper innovatively gave a one-dimensional signal classification method based on generative adversarial networks(GAN).The results showed that using this method significantly improved the classification performance of stroke participants,with an average accuracy of 94.6%.This classification method is helpful to solve complex EMG classification problems.Finally,an active control strategy for rehabilitation robots based on dynamic assistance for was proposed to eliminate trunk compensation.A visual quantitative evaluation system based on a Kinect v2 sensor was introduced to quantify the elimination effect of trunk compensation.Under the action of the platform,the trunk compensation range of stroke participants was effectively eliminated,and the lean-forward angle,trunk rotation angle,and shoulder elevation angle amplitude were reduced by 42.4?0.76%?35.07?1.33%,and 34.67?7.29%.The overall elimination effect was better than that of the predecessors.The total system delay of the platform is 240ms(<300ms),which meets the real-time requirements of the control syste.The platform realizes the automatic detection and elimination of trunk compensation,which helps to improve the effectiveness of rehabilitation training and has potential application value in clinical aspects.