Effect of forearm rotation towards cross-user classification accuracy of forearm gestures / Zinvi Fu … [et al.]

Electromyography (EMG) is a random biological signal that depends on the
electrode placement and the physiology of the individual. Currently, EMG control
is practically limited by this individualistic nature and requires per session training.
This study investigates the EMG signals based on six locations on the lower forearm
during contraction. Gesture classification was performed en-bloc across 20 subjects
without retraining with the objective of determining the most classifiable gestures
based on the similarity of their resultant EMG signals. Principle component
analysis (PCA) and linear discriminant analysis (LDA) were the principal tools for
analysis. The results showed that many gesture pairs could be accurately classified
per channel with accuracies of over 85%. However, classification rates dropped to
unreliable levels when up to nine gestures were classified over the single channels.
The classification results show universal classification based on a common EMG
database is possible without retraining for limited gestures.