Intelligent Perception Recognition of Multi-modal EMG Signals Based on Machine Learning

Surface Electromyography (sEMG) signals directly and objectively reflect the activity of human muscles. As a convenient non-invasive EMG detection method, it is widely used in the field of human action recognition. First, this paper performs active segment detection on the sEMG data collected by the MYO bracelet to extract effective active segments. Subsequently, we extracted five time-domain features from the active segment signal, including the root mean square value, the length of the waveform, the number of zero-crossing points, the mean absolute value, and the maximum-minimum value. Four classifiers, i.e, K nearest neighbor (KNN), support vector machine (SVM), decision tree (DT) and random forest (RF) are used to classify and recognize the extracted sEMG. The highest correct rate is random forest with a value of 82%. Therefore, this paper further extracts the frequency domain characteristics of the signal including the Fourier transform and Willison amplitude. We added 4 models for comparative experiments, including gradient boosting (GB), Gaussian Naive Bayes (NB), linear discriminant analysis (LDA) and logistic regression (LR). The final experimental conclusion show that the effects of the four classifiers have been significantly improved. The best result is SVM for intelligent perception recognition of multi-modal EMG signals, with an accuracy rate of 90% and an F1 score of 0.87.