Analysis Of Multi-features And Application Of Gait Recognition Based On Electromyographic Signals

The EMG signal is generated from the action potential in motor units by muscles shrinking with walking.EMG contains a great deal of physiological information concerning states of the human body,and plays a key role in related muscle system and the neural system.Since it shows the relationship between combination and decomposing in moving patterns and predicts the movement intention,EMG is widely used in diagnosing clinical diseases,recognizing moving patterns and designing new type man-machine interfaces,etc.The typical features of EMG signal which represent every moving pattern are applied to identify the motion states of the human body.This dissertation does a comprehensive research in how to extract more useful features and how to identify gaits more effectively based on EMG signals.All of the issues above are discussed entirely in this dissertation.The main contents are as follows:(1)The algorithm of the wavelet modulus maxima is chosen to preprocess and de-noise EMG signals after analyzing the main noises in the raw data and comparing the common methods.The processed signals conserve intrinsic properties in real EMG,and increase the signal-to-noise ratio(SNR).So,it is better for the subsequent feature extraction and gait recognition.(2)For the instability and chaos in EMG,the Katz algorithm is introduced to calculate the fractal dimensions.By using the fractal features,more detailed information is captured from a higher dimensionality and the whole complexity of the signal is considered.Meanwhile,the average absolute value and the variance are also extracted to build feature vectors for the following gait recognition.(3)The genetic algorithm(GA)is presented to select the best parameters of penalty and kernel function for the support vector machine(SVM)to promote the classifying effects.Based on real human gaits,the initial parameters are set reasonably,so the GA-SVM classifier achieves the optimal accuracy,moreover improves the generalization and stability.(4)For the feature classification of the multi-dimensional non-linear,the qualitative method fails because of the boundary values and cross points.In order to rectify the problem,the improved K-means algorithm is recommended.Based on the density and divergence of samples scattering,the choices of initial clustering centers are optimized,and the isolated noise points are tackled appropriately in improved K-means algorithm.It is proved that the improved K-means algorithm achieves ideal accuracy and accelerates convergence rate.The problems of the gait recognition based on EMG are studied systematically in this dissertation.By applying the wavelet modulus maxima theory,the signal processed is smoother,which has a higher SNR and keeps the nature feature of EMG.The profound understanding of EMG is reached by extracting nonlinear fractal features,as well as the average absolute value and the variance.The GA-SVM classifier is constructed to improve the classifying ability and enhance reliability.The recognition accuracy is prompted and the processing time is shortened by using improved K-means algorithm for classifying the nonlinear features of EMG.These conclusions have significant application importance in EMG,such as the clinical diagnosing,the medical assessment,and the intelligent rehabilitation equipment.