The current study used the transradial amputees as subjects
who are the final user of myoelectric prostheses to
assess the effects of arm position variation on EMG and/
or ACC-MMG pattern-recognition based motion classification
in limb amputees and evaluated the performance
of three proposed solutions in reducing the impact of
arm positions. For amputated arms, the average interposition
error of EMG classification across all the five
arm positions and the five subjects was around 22%
higher than the average intra-position error. This indicates
that the performance of EMG pattern-recognition
based method in classifying movements strongly
depends on arm positions. This dependency is stronger
in intact arm than in amputated arm, which suggests
that the investigations associated with practical use of a
myoelectric prosthesis should be conducted with the
limb amputees as subjects instead of able-body subjects.
Using eight-channel ACC-MMG signals as input of arm
position classifier could achieve an average classification
error as low as 0.7% across five arm positions and five
subjects; even though using two-channel ACC-MMG
signals, the position classification error slightly increased
to 1%. Thus ACC-MMG signals would be very suitable
for the arm position identification. With ACC-MMG
and EMG data as the input signals of arm position
and movement classifier, respectively, the two-stage
cascade classifier could obtain the best performance in attenuating the impact of arm position variation among
three proposed solutions. This suggests that the cascade
classification strategy may be promising for the accurate
and reliable control of EMG pattern-recognition based
prosthetic systems in practical use