A novel method to automatically differentiate forearm movements has been proposed. The electromyography (EMG) signals were recorded from two muscle sites on the forearm in real-time. Two 2-dimensional feature spaces namely the natural logarithm of root-mean-square values (Log (RMS)), and the standard deviations of auto regressive model coefficients (Stdev (AR)) were created. The features were calculated within non-overlapping 0.2 second windows in real-time. The feature spaces were clustered using the fuzzy c-means algorithm . The cluster multiplicities were investigated by five different cluster validity indices. Real-time EMG signal classification was achieved by calculating membership values.
Log (RMS) performed superior to the Stdev (AR) feature space. The silhouette validity index provided the best cluster validity index in this study.
On average, the proposed algorithm classified 4 movements with 92.7± 3.2% and 5 movements with 79.90%±16.8% accuracy. The algorithm also revealed the number of repeatable movements. It can also be adapted to daily variations in individual EMG signals.