Using Forearm Electromyograms to Classify Hand Gestures

Prosthetic hands of increasing capability and sophistication are being built, but how does the user tell the hand what to do? One method is to use the low-level electrical signals associated with forearm muscle movement, or electrogmyograms (EMGs). This paper describes an experiment in which supervised learning, or classification, was used to build a model that decides which of a set of hand gestures was made by a subject based on forearm EMGs. Several techniques were employed to optimize the process. A neurological study was consulted to optimize sensor placement. Several classification algorithms were tried and those with the highest accuracy used. Finally, ANOVA was used to reduce the number of features while maintaining classifier accuracy. The results showed accuracies exceeding 90%, even with a reduced feature set, and that supervised learning has promise as a technique to control a prosthetic hand.