Pattern recognition control applied on surface electromyography (EMG) from the extrinsic hand muscles has shown great promise for the control of powered prosthetics for transradial amputees. The use of limb prostheses is essential for maintaining personal independence and a more effective inclusion in society. However, due to their poor control, imposed by the reduced accuracy of hand movement classification, EMG-driven upper limb prostheses are still not widely used. Hence, post-processing techniques were proposed to reduce the misclassification rates. In this paper, we investigate the effect of two post-processing techniques, namely majority vote and Bayesian fusion, on the performance of EMG-based PR systems when applied on amputees. We measured the effectiveness of a number of time and frequency-based feature extraction methods with different post-processing techniques and various numbers of voting decisions . EMG data was collected from four transradial amputees while imagining seven classes of hand movements. Our results suggested that the recently proposed Time Domain Power Spectral Descriptors can significantly enhance the performance of EMG pattern recognition and that the use of Bayesian fusion can further reduce the error rates upon majority vote with error rates of approximately 5% on average across all amputees.
|Title of host publication||2016 3rd Middle East Conference on Biomedical Engineering (MECMBE)|
|Number of pages||4|
|Publication status||Published - 6 Oct 2016|