A Machine Learning Approach for Muscle Activity Detection

The aim of this contribution is to introduce and validate a powerful approach to detect muscle activations in surface electromyography (sEMG) signals, based on feature extraction and Artificial Neural Network (ANN) classification. The proposed algorithm outperforms the traditional approach, widely used in gait analysis, that is based on a double-threshold statistical detector. The new ANN-based Muscle Activity Detector (ANN-MAD) shows a smaller percentage of erroneous transitions, and a smaller bias and standard deviation with respect to the ground-truth onset detection. The advantages of using the ANN-MAD algorithm are particularly evident for signals with a low or medium signal-to-noise ratio (SNR).