Classification of Human Hand Grasping Force Using sEMG

Abstract

Aiming to classify different hand grasping force levels with the use of prosthetic hands, the electromyography (EMG) signals from the forearm muscles are collected using a commercial surface electromyography (sEMG) sensor. The hand grasping force is recorded using a loadcell. The RMS and the mean frequency (MNF) are used for feature extraction. Both features are extracted using non-overlapping and overlapping windowing techniques. They are applied at different window sizes. SVM, K-NN, and artificial neural networks (ANNs) are used to predict the grasping force levels. The classifiers' performances are evaluated using the classification accuracy and the execution time. The time domain feature obtained the highest accuracies. The K-NN classifier showed the highest classification accuracy compared to the other classifiers. The ANNs produced the shortest execution times among all classifiers. Analysis of Variance is used to show any significance between the classifiers' means accuracies.