Journal of Engineering and Applied Sciences
ISSN: Online 1818-7803ISSN: Print 1816-949x
Abstract
This study presents the implementation of 3 convolutional neural network architectures for the recognition of hand gestures by means of electromyographic signals. The acquisition of signals is done by means of electrodes located in the forearm and the development platform specialized in biomedical signals MySignals HW V2.0 which will be applied a pre-processing of the signal by means of the Wavelet Packet Transform (WPT) for the feature extraction. The architectures that are proposed have as input base to the network the map of features obtained by the wavelet power spectrum with which the database of training and validation was constructed. Finally, in the tests perform in real time, the first architecture reached an accuracy of 93.8325%, the second architecture, reaches a degree of accuracy of 95.8824% and finally, the third architecture reaches an accuracy of 96.4706%. This means that the architecture with the highest accuracy performs better when it comes to recognizing gestures, even with small databases.
How to cite this article:
Natalie Segura Velandia, Robinson Jimenez Moreno and Astrid Rubiano. CNN Architectures for Hand Gesture Recognition using
EMG Signals Throw Wavelet Feature Extraction.
DOI: https://doi.org/10.36478/jeasci.2019.3528.3537
URL: https://www.makhillpublications.co/view-article/1816-949x/jeasci.2019.3528.3537