Motion Intention Recognition Of Lower Limb Prosthesis Based On Fusion Feature And Stratification Strategy

Intelligent lower limb prosthesis refers to that the prosthesis can automatically adjust the joint torque according to the change of the amputee’s walking speed and joint angle,and control the movement of the knee joint and ankle joint,so as to make it close to the natural gait.The intelligent lower limb prosthesis not only reproduces the dynamics and kinematics characteristics of the limb,but also realizes the corresponding movement according to the wearer’s movement intention,maintains the movement stability of the human body,and helps the patient recover the lost functions of standing and walking.The motion intention recognition effect of lower limb prosthesis not only determines the performance of the prosthesis control system.But it will also affect the safety of human beings in sports and work environments.It is one of the core issues that need to be solved in the intelligent control of lower limb prosthesis.The number and types of sensors used in traditional intention recognition methods are large,and the dimension of feature vectors is high.The statistical features used are unstable for short-term samples.In addition,the motor intention recognition data set of intelligent lower limb prosthesis contains the movement patterns and the transition between the modes under different terrain,namely the steady-state mode and the transition mode,which have essential differences in mode and data presentation.If feature extraction and classification of all patterns are carried out directly,the confusion between different patterns will be great,thus affecting the effect of intention recognition.Based on this,a new method for motor intention recognition of intelligent lower limb prosthesis based on fusion feature and layered strategy is proposed in this paper.Firstly,in order to solve the problems such as the large number of sensors and the instability of statistical features to short-term samples,in the second chapter of this paper,only3 D acceleration data and 3D angular velocity data of two inertial measurement units located in the healthy thigh and the lower leg were used to identify the lower limb motion behavior.The method of calculating joint angle is adopted instead of obtaining the change data of knee joint angle from joint angle measuring instrument.Avoided data fusion and other problems,and the dimension of eigenvector is greatly reduced.At the same time,on the basis of not increasing the type and number of sensors,the geometric features and physical features are fused,and the three parameters of acceleration,angular velocity and joint angle are used to excavate the kinematics characteristics of lower limb motion behavior.In terms of feature extraction,the advantage of statistical method in intention identification data processing is retained,and the mean value and variance of physical feature data are extracted to reflect the mean level and dispersion degree of short-term data;Considering the instability of statistical features to shortterm data,the geometric features were extracted to make up for the instability of statistical features and reflect the local rate of change of short-term data.Finally,the motion intention recognition of intelligent lower limb prosthesis was realized.Secondly,in order to reduce the confusion among similar patterns,the hierarchical strategy is proposed in Chapter 3.Based on the feature fusion,the data set is stratified by taking advantage of the differences between classes.The first layer divides the modes into two categories,proposes the method of subtraction between frames to process the data,and classifies them by the difference from the steady states and the transitional states.The second layer divides the samples into two parts according to the classification results of the prediction labels of the first layer,which are trained and identified respectively.Finally,it show that the way could avoid the confusion between the steady states and the transitional states,and the recognition effect is good.Moreover,it provides a compensation mechanism for the results of the first layer,and corrects part of the identification error samples,so as to improve the recognition accuracy and predict the motion intention.