A new approach to mitigate the effect of force variation on pattern recognition for myoelectric control

In myoelectric prosthetic control, the motion classification performance would be decayed if an electromyography (EMG) pattern to be recognized differs significantly from the one used for classifier training. Generally, the training signals are acquired when a subject performs motions with a proper force. In practical use of a myoelectric prosthesis, however, the variation of force levels to do a motion would be inevitable, which will cause a change of EMG patterns. In this study, we proposed a Parallel Classification Strategy consisting of three parallel classifiers created at low, medium, and high force levels, respectively, and designed a regulation to categorize the input signals into corresponding classifiers for pattern recognition. The pilot experimental results demonstrated that the proposed method could enhance the classification accuracy at different force levels with an average classification rate of 98.8%, which was higher than the current method (91.9%). Thus, the proposed method might improve the control performance for present myoelectric prostheses.