Abstract:
In recent times, electromyographic (EMG) signals have emerged as a prominent technique for capturing the electrical activity generated during muscle contractions and rela...Show MoreMetadata
Abstract:
In recent times, electromyographic (EMG) signals have emerged as a prominent technique for capturing the electrical activity generated during muscle contractions and relaxations. These signals can be acquired in a non-invasive manner using surface electrode devices like the 8-channel MYO Armband. This research paper presents a novel approach for classifying EMG signals through a superimposed segmentation technique applied to six hand gestures. The dataset comprises EMG signals acquired from ten patients with varying degrees of hand amputation. To facilitate classification, a continuous Wavelet transform is employed to generate scalogram images, which form multiple datasets with varying resolutions suitable for training a proposed convolutional neural network (CNN). Notably, the experimental results demonstrate an average signal classification accuracy of 91.24% using the one-to-one methodologies. Moreover, these outcomes yield a computationally efficient model with 23,073 trainable parameters, that could serve as the foundation for practical implementation in a device designed for hand gesture classification.
Published in: 2023 VI Congreso Internacional en Inteligencia Ambiental, Ingeniería de Software y Salud Electrónica y Móvil (AmITIC)
Date of Conference: 25-27 October 2023
Date Added to IEEE Xplore: 22 December 2023
ISBN Information: