Accurate sEMG Classification for Hand Gesture Recognition Using Pattern Recognition Techniques

Surface electromyography (sEMG) is a non-invasive method to collect electrical signals produced by muscles during motion. sEMG signals can be used to diagnose diseases, to assist in rehabilitation, control electric prosthesis, or to control robots. These signals are collected using flexible electrodes that are then stored into computers. Afterward, they are further processed to remove noise and extract meaningful features. An important application of sEMG is classification of hand gestures to control electric upper limp prosthesis. These prothesis can either be used to augment a human ability or restore a lost function to increase quality of life. In this paper, we preprocess the sEMG signal, extract hand-crafted features, then apply a proposed classification architecture to classify seven hand gestures achieving an intuitive control to a hand prosthesis. Our proposed model consists of one binary Support Vector Machine (SVM) classifier and one multiclass Artificial Neural Network (ANN) classifier, instead of a single multiclass classifier model. The proposed approach achieved an average improvement of 2% in precision recall and f1 score for all seven gestures in comparison with a single multiclass baseline classifier. The proposed architecture has the ability to classify seven gestures with an average accuracy of 99% using only two electrodes in a short time frame of 200 ms.