Classification of hand movements with electromyogram signals obtained from arm muscles: Generating control signals for hand prosthesis

Electromyography (EMG) signals are outcomes of muscle activities. An EMG signal is read by prickling needle electrode to the muscle or is read non-invasively from the skin surface by placing electrodes on the skin over the target muscle (surface EMG - SEMG). Activities on the biological signals during the times helped to find new methods and technics to both how to use EMG signals and modeling of hand movements. In this study we aim to generate control signals from SEMGs measured from four hand muscles; Extensor carpi radialis, Palmaris longus, Pronator teres, Flexor digitorum superficialis to navigate a prosthesis hand. The SEMGs for five hand movements; finger flexion, wrist flexion, wrist extension, pronation, supination have been acquired. From each muscle (channel), the mean power, peak value of the envelope, the mean frequency of discrete Fourier transform and the estimated frequency by signal subspace method are extracted as features. The different combination of these features have been classified with three different classifiers; linear Bayes classifier, Fisher?s linear discriminant and support vector machine. The classifications have been done for three groups; inter-subject (training and testing data are from the same subject), intra-subject (training and testing data are from different subject), classification of right hand data from the left hand data and vice versa and the performances of the feature combinations and the classifiers have been reported. The highest average performance for inter-subject classification and intra-subject classification have been 99.61% with linear Bayes classifier and 71.8% with linear SVM repectively. The success of the left hand versus right hand (and right versus left) classification with SVM classifier has been at least 36.7% and at most 82.7%. Keywords: Prosthetic hand, EMG, linear Bayes classifier, Fisher?s LDA,SVM