Dynamic factor independent pattern recognition methods for electromyography based upper limb prosthetics

Abstract

An accurate and robust Electromyography (EMG) - Pattern newlineRecognition (PR) design is crucial for the Myoelectric Control System (MCS) to newlineoperate a Myoelectric Prosthetics (MEP). Existing EMG-PR is able to classify newlinevarious hand movements with a high classification accuracy, gt 90% in the static newlinelaboratory setup. Whereas during real-time applications, MEP encounters many newlinedynamic factors that result in reduced classification accuracy, lt 60%. The current newlinechallenges causing the limitation in MEP functionality, during real-time newlineconditions, is categorized into the following dynamic factors in this research: newline(a) Limb position variation: For the same hand movement, the underlying newlinetopography of the muscle fibers may shift, when performed in different limb newlinepositions. (b) Muscle contraction force variation: Muscle force for a hand newlinemovement is decreased or increased, according to the target object. newlineThese muscle force variations produce different muscle activation newlinepatterns for the same movement. (c) Combined forearm orientation and force level newlinevariation: These multiple variations cause inconsistency in the EMG signal pattern newlineformed for classification. (d) Multi-day variation: The performance degradation newlineoccurs in EMG-PR during long-term use, as data acquired for a short period is newlinelimited in information. These changes reduce the similarity of EMG features for newlinethe same hand movement and separability of EMG features for the different hand newlinemovements. newline