Exploring the relation between EMG sampling frequency and hand motion recognition accuracy

Myoelectric control with surface EMG signal has achieved great success in clinics, but only limited to the control of 2-Degrees-of-freedom prosthesis. With the appearance of multiple-channel and high-density EMG system and the advances of pattern recognition technology, it becomes possible to control a multi-degree smart prosthesis using EMG signals. However, it requires high performance EMG systems with high sampling frequency, which impedes the popularity of EMG-based applications. This study aims to explore a way to reduce the cost of EMG system by investigating the effect of sampling rate on gesture recognition accuracy. Two groups of experiments on inner-group and cross-group were designed to evaluate the classification accuracy at different EMG sampling frequency. In comparison with the sampling frequency at 1kHz, a lower sampling frequency at 400 Hz could achieve comparable accuracy, reduced by only 0.43% (KNN) and 0.83% (SVM) with the overall accuracy at 99.40% and 98.67%, respectively. It implies that appropriate reduction of the sampling frequency can be a good choice to balance the cost and performance of a multiple channel EMG system for feature-based hand gesture classification.