Knee Abnormality Prediction based on Power Spectrum Analysis of Surface EMG Signals

Sali Issa¹*, Mahmoud A.M. AlShurafa² and Abdel Rohman Khaled³

Published: April 06, 2022

DOI: 10.55162/MCET.02.034

View Pdf

Abstract  

This article proposes an improved extraction feature for lower limb knee abnormality prediction application using surface EMG raw data. Public UCI dataset is chosen for system evaluation, where each subject was informed to perform three different motions of walking, standing up, and sitting down. In feature extraction, EMG spectrograms are derived using Short Time Fourier Transform (STFT), power spectrum of range 10-250 Hz is chosen, then, the linear coefficients of EMG power spectrum for each frequency bin during time are extracted as a final feature of size 30x2. For system prediction, Convolutional Neural Network (CNN) with other two common machine learning classifiers were constructed. The proposed system experiments proves that EMG signals of Semitendinosus(ST) muscle with CNN classifier produces the highest accuracy of 95%.