Deep learning of EMG signals in hand grip posture identification using time-frequency domain applying STFT.
Juan Pablo Gasca Calderón, Diego Fernando Gonzalez Ruiz, Ruthber Rodríguez Serrezuela
Abstract
Surface electromyographic (sEMG) signals are a noninvasive signal acquisition method that plays a central role in the monitoring of prosthetic devices because they provide information about the motor functions of the human. Therefore, EMG signals must be accurately classified despite signal nonstationarity, sensor noise, involved muscles, and patient peculiarities. This study deals with the classification of hand-grasping postures using sEMG signals acquired by amputee patients. It focuses especially on using the time-frequency domain for feature extraction by applying reduced time Fourier transform (STFT) spectral analysis. The classification model used was a convolutional neural network (CNN), such method was tuned, trained, and evaluated by 2 experiments. The first one called "One by One" where gave an accuracy percentage of 90.84%, 91.05%, and 91.13% for spectrograms of 32x32, 64x64, and 128x128 respectively. The second validation "All by One", obtained a result of 62.28% accuracy for spectrograms of 32x32 pixels.