Improving Myoelectric Pattern Recognition Robustness to Electrode Shift Using Image Processing Techniques and HD-EMG

Abstract

Pattern recognition of myoelectric signals for prosthesis control has been extensively studied in research settings; however, its systems perform poorly in the presence of electrode shift, defined as the movement of surface electrodes with respect to the underlying muscles. In this paper, we present the results of using image processing techniques for gesture recognition in the presence of electrode shifts based on High-Density Electromyography (HD-EMG). Here the instantaneous sample of each EMG channel is represented as a pixel of an image that changes with different movements. In this image, various patterns are recognized as associated with specific gestures. We found that feature extraction based on image processing techniques can improve the accuracy of gesture classification from HD-EMG signals in the presence of the electrode shift.