Light-GBM based signal correction method for surface myoelectropotential measured by multi-channel band-type EMG sensor

Abstract

The aim of this research is to develop a user interface using myoelectrical signals, a type of biological signal. A multi-channel band-type EMG sensor is used to measure this signal. The goal is to make it easier to attach and detach the EMG sensor, while still obtaining sufficiently accurate measurements even if the sensor is placed slightly off the standard position. To achieve this, a Light-GBM based method is proposed to estimate the degree of deviation of the EMG sensor from the standard position. Once the deviation is estimated, a filter is constructed to compensate for the effect of the deviation on the myoelectrical signal. Experimental results using this method show that the sensor misalignment can be estimated with over 90% accuracy, and that correcting the myoelectrical signal based on this information improves joint angle estimation by approximately 13% in terms of root mean square error.