Effect Of Load And Movement Velocity On Muscle Fatigue Classification Based On SEMG Features

As a common phenomenon in daily life,muscle fatigue seriously affects the training efficiency of athletes,the rehabilitation of patients and the work efficiency of workers.Previous studies on muscle fatigue are often limited to fixed environmental variables,but in fact muscle fatigue is generated with varied force conditions.Therefore,the aim of this paper is to explore the variation of EMG features with fatigue and further to classify different muscle fatigue states under complex force conditions.Twelve healthy subjects were recruited in the experiment,and they were required to complete elbow flexion and extension till totally fatigued with different loads(2 kg,4 kg or 6 kg)and motion speeds(period = 4S,8S,or static contraction).Surface electromyogram(sEMG)of biceps brachii muscles during elbow flexion and extension was collected.Statistical methods were used to analyze the variation rules and interfering factors of sEMG features under various motion conditions.The sEMG features sensitive to muscle fatigue were further evaluated and the relationship between muscle fatigue and motion variables was explored.To solve the instability problem of sEMG in dynamic tasks,this paper proposed a method to extract the sEMG data segment in active state,which eliminated the resting sEMG without muscle contractions,and made the extracted sEMG approaching to the sEMG under static contraction.This method improved the stability of sEMG features.Finally,this paper established a general discrimination model across subjects and tasks,and achieved the effective classification of muscle fatigue states under complex tasks.The results show that most of the eight sEMG features selected in this paper can effectively reflect the muscle fatigue state in single and complex motion states,and most of them have no direct relationship with motion variables.Therefore,most of the fatigue features in fixed motion state are also applicable to complex tasks.The result of the discriminant model shows that the final recognition accuracy can be improved by considering the motion state during model training.After the preliminary classification,the voting method is introduced to effectively improve the performance of the model,and finally the comprehensive classification accuracy is 83.3%,which proves the feasibility of fatigue state classification by sEMG under complex tasks.