Analysis Of Human Upper-limb Motion Intention Based On SEMG Signal

With the development of technology,robots are more and more deeply into all fields of human society,and play an increasingly important role.Whether the exoskeleton robot that providing assistance or the rehabilitation robot that providing rehabilitation training for patients,which all need to interact with human body in real-time,which are called humancomputer interaction robot.However,the traditional human-computer interaction algorithm makes the robot unable to accurately understand the motion intention of the human body,so it is easy to lead to muscle fatigue,injury and other situations,which limits the development and application of human-computer interaction robot.It can be seen that in the process of human-computer interaction,the key to realize human-computer interaction is to understand human motion intention through human motion intention analysis.Therefore,how to realize the quantitative,accurate and real-time analysis of human motion intention becomes a key problem to be solved urgently.The main contents of this paper are as follows:First of all,the generation mechanism of surface electromyography(sEMG)is analyzed,and the pre-processing of sEMG is carried out;considering the non-synchronization in timedomain among s EMG,angle signal and moment signal caused by electromechanical delay,the second-order recursive filter is used to compensate the time-delay;the gradient boosting tree(GBT)model is introduced to calculate the contribution rate of the input channel of sEMG,and the channel of sEMG is selected;the Inverse Dynamics in Opensim software is used to establish and analyze the dynamic model of human upper limb system based on Lagrange equation,then changes the model parameters for different human bodies based on Scale Tool,which lays the foundation for quantitative analysis of human motion intention.Secondly,a musculoskeletal model(MSM)was proposed to analyze the movement of the patients of hemilateral dyskinesia.Based on the physiological structure of human body and the health musculoskeletal model(HMSM),the similarities and differences of physiological parameters and structures between the health side and the disabled side were discussed.Based on the above discussion,the disabled musculoskeletal model(DMSM)was established,then the image method is used to mirror some physiological parameters and structures of the health side to DMSM,then the genetic optimization algorithm is used to re-optimize the remaining physiological parameters in HMSM;finally,take root mean square error(RMSE)and prediction time as indicators to verify the effectiveness of the proposed GBT and DMSM method through experiments.Thirdly,a structure of multiscale convolutional neural network(MSCNN)is proposed to meet the needs of the analysis of motion intention of health people's upper limb,by using its convolution structure of multi-scale characteristics to fit the characteristics of sEMG in different scales,improving the accuracy and stability of fitting;by using the convolution kernel calculation method characteristics to reduce the information redundancy caused by the multi-input channels of sEMG,as well as the impact of the increase in the amount of calculation and the stability of the model;finally,the effectiveness of the proposed MSCNN method is verified by experiments based on RMSE and the prediction time.Finally,a human motion intention analysis system is built,including information acquisition module,algorithm analysis module and online display module.On this basis,an experimental scheme of human upper limb motion intention analysis is designed,which adopts the experimental paradigm of keeping the upper arm vertical to the ground and lifting the lower arm at constant speed to verify the two algorithms of human upper limb motion analysis proposed in this paper.