Characterization of signal features for real-time sEMG onset detection

Objective: Onset detection of surface electromyography (sEMG) is useful in sEMG-controlled systems because it can reduce the response delay to user intention. Various data processing techniques have been proposed for onset detection; however, the properties of onset features have not been investigated and compared because qualifying such detection is not simple.Methods: This study clearly defines false detection, detection failure, and delay to quantitatively characterize the onset features. Then, the performances of nine features and their changes caused by the size of the sliding window are obtained; the sliding window is needed to calculate the feature value.Results: Among the features, the variance of the signal showed the largest delay and lowest false detection. The delay from waveform length was the lowest, but its possibility of false detection was highest. As the window size increased, the delay reduced while the correlation between the window size and false detection differed over the features. Additionally, a new concept called label-shift was developed to modify the characteristics of the fea-tures. A positive label-shift increased the delay and decreased the possibility of false detection.Conclusions: The delay and false detection significantly vary over features, window size, and label-shift, and thus researchers need to consider these factors to design sEMG onset detection system. Significance: The properties revealed in this study can be used to select the optimal features, window size, and label-shift for a target application.