Image set preparation: A platform to prepare a myoelectric signal to train a CNN

Abstract

Derived from the good performance in the classification of surface Electromyography signals using CNN for its application in prosthetics, rehabilitation, and medicine, we present a platform that, from a surface Electromyography, performs the necessary digital processing to generate an image database to train a Convolutional Neural Network. This platform requires inputting the protocol parameters under which the myoelectric signal was acquired. In addition, it allows selection among four groups of Time-Domain features and four types of images that have shown good performance (above 90%) in the current literature. The platform generates images in separate folders for each movement according to the selected parameters. This work offers a valuable tool in classification using surface Electromyography and Convolutional Neural Networks, enabling more efficient customization and optimization of training processes.

Keywords

• CNN network
• Gesture classification
• Image processing
• sEMG