Feature Analysis Of Lower Limb SEMG And Research On Gait Recognition Algorithm

In recent years,intelligent bionic legs have become a subject of much attention in robotics,rehabilitation medicine and biomedicine,the core of the algorithm of wearable bionic legs is the study of gait recognition.There are some existing research on gait recognition:the problem of intention recognition caused by the lack of biological information interaction,the lack of walking research problem which is closer to the reality scene,the low precision of gait recognition algorithm based on sEMG and the robustness of algorithm in complex environment.In this paper,the following research work has been carried out:(1)Reasonable designing data acquisition experiment and preprocessing method.In the acquisition of experimental data,this paper reasonably designs the data acquisition experiment of simple walking,walking at non-synchronous speed and walking under cognitive task,and analyzes and studies the walking process with the sEMG as the signaling source.In signal preprocessing,this paper uses comb filter and band-pass filter to filter out the ECG noise in the sEMG and the power frequency noise of the hardware,which can filter out the signal noise more effectively than the traditional method of empirical modal decomposition and denoising.(2)Research and analysis of feature extraction.In this study,four kinds of features extraction methods of sEMG,including time,frequency,time/frequency domain features based on wavelet analysis and entropy,are verified by scatter plot of features,statistical analysis and classifier,indicated that most time domain features are superfluity and redundancy.The experiment results show that Slope sign change(SSC),Waveform length(WL),Wilson Amplitude(Wamp),Logarithm of variance(logvar)and absolute mean value(DB7-MAV)based on DB7 wavelet decomposition coefficients and fuzzy approximate entropy(FuzzyApEn)have good performance in gait phase detection.(3)Research on gait analysis and gait recognition algorithm.In this paper,the walking data and features of different environments are compared and analyzed,and the analysis shows that too slow or too fast will lead to the blurring of the classification boundary in the subdivision stage of the stance phase and the swing phase,resulting in the increase of the steady-state error.While the walking performance under cognitive task has a slight influence on each stage of gait,especially in the transition period of the support phase and the swing phase of the boundary blur,resulting in the increase of transition error.From the above analysis,this paper presents a linear discriminant analysis based on kernel method optimization,which effectively reduces the influence of pace and cognitive task on gait recognition.On this basis,according to the characteristics of"good and different"of the selected features,the method of nonlinear integrated learning Stacking is used to effectively improve the overall recognition accuracy and the generalization ability of the algorithm.So that the gait recognition accuracy under simple walking reaches 94.7%,and the gait recognition accuracy under complex walking reaches 92.5%.The recognition speed is about 85ms,which is 3%compared with the traditional algorithm,and the robustness of the algorithm is better,which satisfies the real-time requirement of gait recognition algorithm,and the research results are of great value to the application of lower extremity pattern recognition.