Abstract:
This work presents the development of a robust hand gesture classification system based
on surface Electromyography (SEMG) signals, designed for real-world applications. The
project utilized data collected from subjects wearing electrodes on their hands, with initial
investigations conducted using the wrist electrode data from GrabMayo dataset.Later on used
statistical and anatomical knowledge to find the best electrode placement positions so that
amount of data needed and complexity of model are reduced. Preprocessing techniques were
employed to convert the raw SEMG signals into mel spectrogram images, alongside extraction
of time domain features. These features were then utilized to train and fine-tune machine
learning models including Support Vector Machines (SVM), Random Forest, and K-Nearest
Neighbors (KNN). Subsequently, the study explored deep learning approaches, utilizing both
pretrained neural networks and a custom Convolutional Neural Network (CNN) architecture.
The investigation aimed to increase classification accuracy while reducing model complexity
to facilitate implementation on wearable devices. Comparative analysis of the machine
learning and deep learning models was performed using performance metrics. Additionally,
a robust hand gesture classifier model with small size was developed with good accuracy
and low latency in prediction. The findings of this investigation can contribute to the
advancement of gesture recognition systems, offering insights into effective deep learning
approaches for SEMG-based hand gesture classification in real-world scenarios.
