Recognizing Subjects Who are Learned How to Write with Foot From Unlearned Subjects Using EMG Signals

In this paper we report the preliminary results of recognition of learned subjects from unlearned ones during foot writing process using electromyogram (EMG) signals recorded from thigh and shank muscles. For proof of idea, three subjects were asked to write seven letters with foot. We recorded and analyzed the data to study the learning process in five sessions. Since previous studies have shown that pressure of pen and stiffness of hand are inversely related to the learning level, we considered the pressure applied by a magnetic pen on a digital tablet during foot writing as one of the main features to represent the learning level. Pressure analysis demonstrated that the pressure on the tablet is decreased during successive task accomplishments. Statistical analysis of pressure indicates that the fourth day can be considered as the frontier of the learning process. The EMG signal was also recorded from eight leg muscles and for each of them 28 features were extracted. Several classification methods including Support Vector Machine (SVM), Linear Classifier, Naive Bayes and K Nearest Neighbor (KNN) were used in order to classify the recorded data. With 10 superior features chosen by Sequential Floating Forward Selection (SFFS) algorithm for each classifier, the accuracy of corresponding classifiers was in the range of 72%-95%. Moreover, we found out the SVM classifier, and the two Tibialis Anterior (TA) and Medial Gastrocnemius (MG) muscles could distinguish between learned and unlearned subjects most properly. The accuracy of features Mean frequency (MNF), Modified Mean Absolute Value (NMAV2) and Third spectral moment (SM3) belonging to TA and Willison Amplitude (WAMP) and Root Mean Square (RMS) belonging to MG was 90 and 85 percent, respectively. This preliminary study suggests that EMG signals can be effectively used to determine when the learning procedure is converging (or starting to converge) to its steady and ultimate level.