In most reported studies of EMG pattern recognition
systems for multifunctional prosthesis control, subjects
generally took a seated position with a tested arm resting
on a plate surface such as chair arm or table and multichannel
EMG signals were acquired with a number of
surface electrodes placed on either the muscles of forearm
and hand for an able-bodied subject or the residual
muscles for an upper-limb amputee. One portion of the
acquired EMG data was used to train a classifier and
then remaining portion was loaded into the trained classifier
for calculating the offline classification accuracy in identifying a number of arm and hand movements
[4-18,21-23]. With this experimental setting mode, high
classification accuracies were often achieved since the
training and testing EMG data could be consistently
recorded in a constant position of the tested arm. However,
this procedure would be different from the clinical
application of a multifunctional myoelectric prosthesis,
where the user’s arm position varies when he/she is
going to activities of daily living.