Electromyography Signal Classification using Convolution Neural Network Architecture for Bionic Arm High Level Control

Deep Learning is a trending control method in recent applications and studies. It is used in complex classifiers where traditional condition classifiers fall short in complexity and versatility. Many previous attempts use Deep Neural Networks to classify Electromyography (EMG) signals; however, most required high computational power to achieve acceptable real-time operation. By implementing a convolution neural network (CNN), we can have both high accuracy and real-time operation using off-the-shelf microprocessor boards that can be easily embedded in a bionic limb for seamless control. In this project, we use a commercial EMG sensor (Myo Armband) to gather data needed for training the CNN for reliable classification of 14 different bionic arm poses. Progress done in this study can be used for repeatable and reliable control over bionic limbs (tests were on a bionic arm for forearm amputees). It can be also used as a control interface by gesture recognition as a safe interaction medium between human and machines.