Hand Gesture Classification using Deep learning and CWT images based on multi-channel surface EMG signals

Nowadays, electromyographic signals are one of the most widely used techniques to acquire the electrical activity produced by muscle contractions and relaxations, these signals can be measured in a non-invasive way using surface electrodes devices such as the 8-channel MYO Armband. This paper presents a method for the classification of EMG signals using a superimposed segmentation of the EMG signal of 6 hand gestures acquired from 10 patients with different levels of hand amputation, to subsequently apply the continuous Wavelet transform generating scalogram images forming several datasets with different resolutions suitable for the training of the proposed convolutional neural network. In this experiment, we achieved an average signal classification accuracy of 85.70% and 94.49% in the all-to-one and one-to-one methodologies, respectively. The results generate a model with low computational cost, which can be the basis for actual implementation in a device to classify hand gestures.