Estimation of Finger Joint Movement based on Electromyography Signal using Long Short-Term Memory

The presence of hand plays a vital role. Without a hand, humans experience difficulties in their activities. As a result, several solutions have emerged to overcome this problem, especially finger movement regression using electromyography (EMG) signals for specific movements such as extension/flexion. Therefore, this study proposes a regression task on surface EMG (sEMG) collected from double Myo-Armband on finger movements. This experiment uses feature extraction of Mean Absolute Value (MAV) and Root Mean Square (RMS). Dimensionality reduction is then conducted to speed up the regression process using Principle Component Analysis (PCA), Independent Component Analysis (ICA), Non-Matrix Factorization (NMF), and Linear Discriminant Analysis (LDA). The last is estimating angle finger joint using Long Short-Term Memory (LSTM). The results show that the best performance is in the RMS and PCA features with an R-Square value of 0.874, and ICA and RMS perform the fastest time with an R-Square value of 0.871.