Electromyography (EMG) Based Finger Movement Detection

Abstract

One fundamental component of much modern human-machine interaction (HCI) devices is Myoelectric control systems which is a system that receives the Electromyography(EMG) signal originated from muscle movement. Much research has focused on determining the best general structure of the control system for a given application where the same element choices are used for all subjects. However, due to the nature of the signal and human body, the best structure may be subject-specific. The primary aim of this research can be categorized into two major areas. 1)Movement extraction and movement duration detection from a recorded set of moves. 2) Subject-specific selection of classification system elements (i.e., feature set, classifier, window characteristics, dimensionality reduction method) for individual finger movement detection. In this study, we focus on individual finger movements, therefore, two movement sets for each finger were tested: one where each finger is closed for half a second and open for half a second and the other with the duration of one second closed finger one-second open finger. Myoelectric data were collected from the forearm muscles of 27 years old female subject using a single channel Epidermal Electric System (EES). We performed three sets of tests on our subject; giving us three data sets to study and develop a model for. These data were first got prepossessed for movement extraction then used to train and test a series of classification systems, each consist of a different combination of system element choices. We introduce a novel model for movement extraction from an unfiltered EMG signal and achieve an average accuracy of 88% for our five class finger movement classification. Additionally, we show the effect of Principal Component Analysis on Classification, as well as multi-layer classification of EMG signals.