A Novel Combination Method of a Convolutional Neural Network and Energy Operators for the Detection of Change-Points in Electromyographic Signals

Currently, neural network algorithms based on time-domain features are used for change-point detection problems, and they have proven to be effective. However, due the instability of human biosignals, establishing a training dataset with labels is difficult. For supervised learning methods, wherein parameters updated small sample set through feed-forward mechanism, it difficult ascertain degree which performance trained corresponds overfitting upon was trained. To this end, paper attempted directly replace in convolutional that need by training. A method combination Teager–Kaiser energy operator (TKEO) proposed. We tested proposed simulated EMG data different signal-to-noise ratios real labels, respectively. Compared multiple had significant advantages terms reliability, accuracy, computational speed. Furthermore, does not require any prior knowledge about signal, lending itself flexible adaptable application. It may promising alternative solving problems.