Research On Lower Limb Motion State Recognition Based On Multi-source Information

Wars,diseases and natural disasters have caused a large number of people to become physically disabled.Prosthetics can help residual limbs restore certain limb functions and enable them to integrate into society as ordinary people.The traditional mechanical prosthetic gait is stiff,and different road conditions require manual adjustment by the user,so that the prosthetic wearer is unnaturally walking and easily feels tired.The intelligent prosthesis equipped with microprocessor and sensor can adaptively adjust the prosthetic gait according to the change of walking environment,making walking more natural and greatly improving maneuverability,comfort and coordination.The key issue facing intelligent prosthetic research is how to realize the information interaction and intuitive control between the prosthetic and the wearer.In order to enable the prosthesis wearer to achieve natural walking,the prosthetic system needs to obtain the user's walking intention through the sensing system,and provides the user with the corresponding control strategy and action assistance based on the basis,thereby realizing the accurate identification of the prosthetic wearer's walking intention.It is the premise of adaptive control of intelligent prosthesis.Information commonly used to estimate motion intent primarily includes kinematics,dynamics,and myoelectric information.Considering that a single signal is not enough to provide all the information to judge the gait pattern,this paper builds a multi-mode information acquisition system to obtain gait key information,and explores the performance of three different recognition algorithms,and establishes a prediction model to achieve lower limb movement.Preliminary forecast.Firstly,the multi-mode signal acquisition system is used to obtain the EMG signal of the lower limbs reflecting the active motion information,the acceleration signal of the spatial position change,the pressure signal of the foot force condition,and the knee joint angle signal that visually reflects the change.The wavelet threshold denoising is used to denoise the EMG signal,and the acceleration,the plantar pressure and the knee joint angle are smoothed,and the processed signals are extracted separately.In multi-motion pattern recognition,the extracted features are filtered by Relief-F,which reduces system consumption while ensuring recognition accuracy.Then LDA,SVM and LM-BP algorithms are used to establish the recognition model of multi-motion mode,and the training and simulation are carried out.By comparing the recognition accuracy and time consumption,it is found that LDA has the best comprehensive performance among the three types of algorithms.In addition,the RBF neural network is introduced in this paper to establish a human lower limb motion prediction model with knee joint angle signal as input information.The experimental results show that the results of the lower limb motion state prediction model established by RBF neural network are more accurate,which is almost consistent with the actual motion trend.Can provide a reference for prosthetic control.