sEMG-based decoding of human intentions robust to the changes of electrode positions

In this paper, a novel sEMG decoder was proposed for estimating the human intention. When the position of the electrode changes, most of the decoder cannot be used as it is and needs to be retrained with new data from changed electrode positions. In retraining, old training results are often discarded and a new decoder is trained with only the data gathered after the changes. Naturally the decoding performance is hard to be maintained. Unlike these conventional decoders, the new decoder does not discard the old decoder and use it after a small modification to compensate the changes so that the decoding performance can be maintained after the changes. The core of this robustness is a newly proposed feature extractor named as class-augmented independent component analysis, which is composed of the ICA to extract independent components as new features and a scheme to reorder and match the new features with the old decoder features. Experiments confirm that the proposed method decodes human intentions from sEMG with a high accuracy and this performance can be maintained even the electrode position changes.