Movement Pattern Recognition Based On Surface Electromyography Of Thigh Stump

Wearing lower-limb prostheses is the only way for amputees to come back to daily life.In order to realize the flexible control of prosthesis,the movement intention of disabled people must be identified quickly and accurately.Surface electromyography(SEMG)information can reflect the movement intention of human body and is generated before the real human movement.So compared with the motion information,SEMG signals has a remarkable advantage in movement pattern recognition.At present,the research of SEMG pattern recognition is mainly concentrated in healthy people.However,since limb amputation,the SEMG signals of amputees are different from healthy people,so it is important to use the stump SEMG information to identify different movement modes.In this paper,SEMG signals collected from muscles of lower-limb amputees are used for identifying five different movement patterns.The recognition results can be used for lower-limb prosthesis control.The research contents are as follows:Firstly,impact of lower limb amputation is analyzed,muscles and method of collecting SEMG signals are determined.Specific to the problem of present method,socket-sensor system is designed to collect SEMG signals.Secondly,by analyzing the characteristics of SEMG signals in different movement modes,a multi-phase detection algorithm is proposed based on moving windows,which realize the extraction of multiple data segments in a moving period.Features of multiple data segments are extracted,and feature vector is constructed.Thirdly,Specific to the misleading and redundant features,BP neural network method based on genetic algorithm,partial least square method based on genetic algorithm,and BP neural network method based on mean impact value are applied to dimension reduction.According to the dimension reduction results,the optimal feature vector is selected.After dimension reduction,feature numbers are obviously decreased,and the recognition accuracy rate has improved.Finally,the optimal parameters and data length of each data segment are selected,and random forest algorithm is applied to the movement pattern recognition of multiple data segments.According to the recognition result,a periodic pattern recognition method of binary tree is proposed based on random forest,which realizes the periodic pattern recognition of five movement patterns.The simulation result indicates that this method can not only improve the recognition accuracy rate,but also improve the real-time performance of prosthesis control.