Electromyography Signal Recognition Method Based On Topological Data Analysis

Electromyography(EMG)signal is a bioelectric signal accompanied during muscle contraction,and it is an important method to detect the activity of muscle groups.By collecting and analyzing the EMG signals of the muscles near the hand,it can be used to identify the type or intensity of hand movements,realize the classification of hand movements,and then operate the external machine.In this work,we propose a method to classify the EMG signals generated by hand movements using topological features(Persistent entropy,Wasserstein amplitude,Landscape amplitude and Betti amplitude).First,the original EMG signals are converted into point cloud data embedded in Euclidean space using the time-delay embedding method.Then,a multiscale topological space is constructed on the point cloud data,and four topological features are extracted by calculating persistent homology and vectorizing persistence diagrams.Finally,these features are input into a classifier for hand movement classification.This paper explores four main questions:(1)How to choose the optimal embedding delay and embedding dimension to transform time series into point cloud data;(2)Which dimension of homology(H0,H1)contributes the most to the classification of EMG signals;(3)Among the four selected topological features(Persistent Entropy,Wasserstein Amplitude,Landscape Amplitude,and Betti Amplitude),which is most effective for classifying EMG signals;(4)Whether topological features can help identify and classify EMG signals or whether the "shape" feature of EMG signals can serve as the basis for classification,which is the most critical issue addressed in this paper.The main results of this paper include:(1)Generally,the higher the embedding dimension,the higher the classification accuracy,while the embedding delay has little effect on the classification accuracy;(2)In the classification of hand movements,the topological features calculated from the first dimension of the homology group(H0)are more effective than those of other dimensions;(3)Betti amplitude is a more stable and effective feature than other types of topological features;(4)Topological features are effective for classification of EMG signals and outperform other time-domain features in experiments.In addition to the main findings,we used Betti curves to visualize the topological patterns of EMG signals generated by hand movements.