IMPROVEMENT OF THE GUILIN-HILLS MYOELECTRIC SIGNAL CLASSIFIER FOR PROSTHESES CONTROL UNDER THE INFLUENCE OF FATIGUE

摘要

Control of upper limb prostheses can be realized via therndetection, processing and classification of myoelectric signals.rnPattern recognition techniques enable users to interactrnwith computers by gestures or hand movements. Thernprime application of our work are hand prostheses controlledrnby myoelectric signals. The developed techniquesrnand the work is not limited to prostheses control and isrnequally applicable for the control of exosceletons, gamesrnand other computer devices. With our established classificationrnalgorithm, the Guilin-Hills Selection Method, werncan distinguish up to nine hand positions. The method isrnbased on statistical cluster analysis and is superior to thernpreviously employed neural-net approach. Since musclernfatigue affects the accuracy of the classification algorithms,rnthe probability density functions of time-domain featuresrnare no longer normally distributed as previously assumed.rnIn this paper, we present an enhancement of our classificationrnalgorithm yielding better classification results underrnthe influence of fatigue. The new method is a kernel-basedrnapproximation of the features probability density function,rnleading to fewer class overlap and better classification results.