Hand and Lower Arm Movements Classification Using Deep ANN and sEMG

In this paper, we develop an Artificial Neural Networks (ANN) based softmax classifier mated to ANN sparse autoencoders to classify human's hand and lower arm movements in healthy participants utilizing muscles' Electromyography (EMG) signal datasets from [1]. We preprocess the data for the classifier in three main ways; filtering, replacing the data with its equivalent RMS value, and finally, dimension reduction using two stacked autoencoders. The classifier achieved a classification accuracy of more than 99% for the first data set with 10 moves, and more than 93% for the second data set with 26 moves. The average classification time for a single movement was 2.1 microseconds. Comparing these results to [1] we find over 7.6% classification accuracy improvement and more than 23% increase in classification speed; both improvements are essential in developing controllable prosthetic limbs.