DECODING OF DIFFERENT HAND MOTIONS

Abstract

Electromyography (EMG) is an electrical signal generated due to contraction of muscles of the body. EMG signal recorded directly from the muscles by utilizing needle wire electrodes is known as intramuscular EMG signal. Intramuscular EMG signals can be used to design control for prosthetic limbs. Pattern recognition-based schemes employ intramuscular EMG signals to design control for prosthetics limbs with multiple degree of freedoms (DOF). Feature extraction plays an important role in the performance of the designed system. In this study, intramuscular EMG signals were recorded from 5 healthy subjects, and various time domain features have been investigated to evaluate the effect of various features on the performance of myoelectric control system. The results shows that by increasing number of features the performance of the system increases drastically. The mean classification accuracy (MCA) averaged across all subjects increased from 40.62% to 83.54%. Statistical analysis showed that, the MCA of feature set with single feature is statistically significant different from the MCA of all other feature sets with multiple features (P-value = 0). The subject wise variability in performance of the system has also been observed. Statistical analysis showed that, subject 4 achieved highest average CA of 78.80% among other subjects and it is statistically significant different from the CA of subject 1 (P-value = 0.0080) and subject 2 (P-value = 0.0432). whereas, no statistically significant difference has been observed between the CA of subject 4 with CA of subject 3 (P-value = 0.9191) and subject 5 (P-value = 0.4161). The presented results can be used to design a pattern recognition based myoelectric control for upper limb prostheses devices with better and intuitive control.