Abstract:
In biomechatronics systems, for efficient and consistent control, the utili sation of muscle activation signals in the control loop is essential.. Motion
categorization utilising electromyography (EMG) signals that represent mus cle activity is one of the approaches utilised for this goal. It’s tough to classify
these signals with varying amplitude and frequency. EMG signal properties,
on the other hand, alter throughout time based on the person, task, and du ration. For movement classification, a variety of artificial intelligence-based
algorithms are applied. Machine learning and deep learning are techniques
used these days. In this paper we have applied different machine learning
techniques as well as deep learning techniques to classify EMG signals. We
have used dataset “EMG data for gesture dataset” taken from UCI repository
with 6 gestures. We have used novel methods such as 1-D CNN, MobileNets
and Xception model that are more reliable and efficient techniques as com pared to classical machine learning techniques. We have achieved up to 98-99
percent of test accuracy on MobileNets by classifying gestures. Comparing
all the different techniques we came at the conclusion that deep learning
architectures are best in terms of classification of EMG signals.
