Research On Decomposition Algorithm Of Surface Electromyographic Signal Based On CKC

Human movement information can be widely used in human-computer interaction,medical and health fields.Electrophysiological signals containing movement information can be generally divided into three categories: electromyographic signals,electroencephalogram and peripheral nerve signals.Among them,the acquisition of electromyographic signal is relatively simple and convenient to process,and the generation of electromyographic signal is ahead of the occurrence of motion,which is widely used in the extraction of human motion information.According to different acquisition methods,electromyographic signals can be divided into intramuscular electromyographic signals and surface electromyographic signals.Compared with the intramuscular electromyographic signal,surface electromyographic signal acquisition is more simple and non-invasive,and has a wider range of applications.In the process of accurate acquisition of motor intention and diagnosis of neuromuscular diseases,it is necessary to further decompress surface electromyographic signals to obtain the time series information of motor unit release.At present,the decomposition methods are represented by Convolution Kernel Compensation algorithm.However,the decomposition speed of CKC algorithm is too slow and can not meet the real-time application.In response to these issues,the specific research content of this thesis is:(1)In off-line decomposition methods,based on the compensation algorithm for convolution kernels and the compensation algorithm for gradient convolution kernels,a fast gradient convolution kernel compensation algorithm is proposed.This algorithm is used to extract the time series of motion unit release in off-line semg signal.In gradient iteration,this model not only considers the observed value at the current time,but also comprehensively considers the influence of all previous observed values at the current time,so as to accelerate the gradient convergence speed and improve the decomposition efficiency of surface emg signal.In the simulation experiments of surface emg,it is found that under different signal-to-noise ratio conditions,the decomposition speed of the proposed algorithm is increased by 3-4 times compared with g CKC algorithm,which solves the problem of slow off-line decomposition speed and proposes a more efficient method for extracting the motion information of surface emg.(2)In on-line decomposition methods,the problem of difficulty in real-time decomposition of surface electromyography signals using off-line decomposition methods and inability to meet the requirements of real-time response scenarios is addressed.Based on the off-line decomposition method,sliding window sliding EMG is adopted.The on-line decomposition method includes two processes: off-line pre-training and online sliding window.The off-line pre-training uses the fast gradient convolution kernel compensation algorithm to compute the cross-correlation vector and further clusters the time series of the moving units and inputs the results into the on-line decomposition process.When calculating the cross-correlation vector through on-line decomposition,the complex iteration is no longer carried out,but the discharge time generated by the same moving unit is merged into the same set.The sliding window adaptively updates the cross-correlation matrix and cross-correlation vector according to the decomposition results,and finally estimates the time series of the sending of moving units.Compared with the off-line decomposition method,when the sliding window size of the on-line decomposition method is 3 seconds,the surface emg decomposition can be completed in milliseconds,which meets the requirements of real-time decomposition and satisfies some real-time response requirements.