A novel supervised feature extraction for decoding sEMG signals robust to the sensor positions

In this paper, we proposed a novel supervised feature extractor named as class-augmented independent component analysis (CA-ICA) whose performance can be maintained even after the input variables are varied, only if new input variables are still linear combinations of the same independent sources as old input variables were. This property can be useful in implementing an sEMG decoder robust to the position changes of sensors (electrodes), since the electrodes attached at a position on human skin is not easy to be maintained for a long time. Experiments show that the sEMG decoder with the proposed method decodes human intentions from sEMG with a high accuracy and this performance is maintained even if the electrode position changes.