Bi-directional LSTM for Monitoring Biceps Brachii Muscle Activity of Healthy Subjects Using sEMG Signals

In recent decades, EMG (Electromyogram) pattern recognition systems focus primarily on feature engineering techniques. Recently, feature learning has demonstrated better recognition accuracy in many research fields than the well-established hand-crafted features. The proposed research work demonstrates designing an EMG pattern recognition system for characterizing slow and fast muscle activities of 40 healthy subjects and monitoring their muscle functions. The analysis used a network of Bidirectional long-short-term memory (Bi-LSTM) with five layers. Feature Learning using Bi-LSTM is performed to monitor muscle activities during continuous forearm extension 5 times with slow and fast movements, using a medium-sized dataset extracted from 40 healthy male subjects. Simulation experiments were conducted with and without cross-validation, which reported an accuracy of 92.86% and 93.75%, respectively, suggesting that the proposed model can be developed as a generalised model for EMG analysis.