Research On Recognition Of Movement Intent And Fatigue Estimation In Rehabilitation Training Of Lower Limbs

In recent years,cerebrovascular diseases such as"cerebral stroke"have caused a large number of patients with limb movement dysfunction,which has seriously increased the burden on families and society.At present,a mainstream method is to train the affected limb through a rehabilitation robot system.However,the existing training systems have the problem that the training mode cannot be adjusted in real time according to the patient's movement intention and fatigue state,which reduces the patient's active participation and easily causes secondary damage.In order to solve this problem,this thesis collects surface electromyography?sEMG?and electrocardiogram?ECG?in the process of the lower limb training to conduct research on movement intention recognition and fatigue estimation.The offline model of movement intention recognition based on sEMG and the offline model of fatigue estimation based on the fusion of sEMG and ECG features are constructed respectively.Furthermore,based on the lower limb rehabilitation robot platform,the online application of both offline models is realized.The main work is as follows:Firstly,based on the analysis of the main movements and muscles of the lower limb joints and the needs of patients with movement dysfunction,6 groups of lower limb intention movements and 3 types of fatigue states are selected.And then,the experimental schemes of sEMG and ECG acquisition for movement intention recognition and fatigue estimation are designed.In order to remove the noise in the original signal,a linear filtering method based on wavelet threshold is designed,and a better denoising effect than wavelet threshold and filter is achieved.Secondly,an offline model for lower limb movement intention recognition is constructed.For the action segment extraction of the sEMG,the start and end points of the action are detected by calculating the sample entropy feature of the signal sliding window framing,which improves the accuracy of action segment extraction.For intention features,7 kinds of time domain,frequency domain and time-frequency domain features including energy entropy are extracted from sEMG,and principal component analysis is used to reduce the redundancy between features.In order to accurately classify intention features,a movement intention classification method of probabilistic neural network based on fruit fly optimization algorithm is proposed.Through cross-validation of single features and combined features,the best combination of intention features is found as ZC+WL+MF+energy entropy.The recognition rate can reach 93.33%,realizing the effective recognition of 6 groups of movements.Thirdly,an offline model of lower limb fatigue estimation is constructed.Aiming at the problem that training fatigue cannot be accurately detected by pure sEMG signals,sEMG and synchronous ECG signals are combined to perform fatigue estimation research,and 12 kinds of time domain,frequency domain and time-frequency domain features including EMG frequency band value and ECG_MEAN are extracted.In order to achieve accurate classification of fatigue features,a fatigue classification method of multi-class support vector machine based on feature fusion coefficients of sEMG and ECG optimized by particle swarm optimization is proposed.Through cross-validation of single features and combined features,the best fatigue fusion feature is found as WL+MPF+band value+ECG_MEAN+ECG_LF.The recognition rate can reach 98.33%,which is 7.66%higher than that of the sEMG features alone,realizing the effective recognition of 3 types of fatigue states.Finally,in order to verify the online application effect of the offline model of movement intention recognition and fatigue estimation,a passive training experiment based on fatigue estimation is designed,and the online recognition rate of fatigue states is85%.At the same time,an active training experiment based on movement intention recognition and fatigue estimation is designed,and the online recognition rates of fatigue states and movement intentions are 87.5%and 81.67%respectively.The experimental results show that the method can effectively improve the patient's active participation,reduce the rehabilitation fatigue and extend the training time.