A wearable, real-time sEMG gesture classifier based on E-tattoo and CDF-CNN for prosthetic control

Abstract

Bionic prosthetic hands are essential for the disabled to deal with most affairs of life independently. The surface electromyography(sEMG) is considered as a developing vigorous solution to realize the artificial limb system. However, most of the present devices are irritating to the skin, high cost and power energy, complexity, unnatural, which limits their promotion. This paper proposed a real-time bionic mechanical arm control system, based on the large-area flexible electronic tattoo(E-tattoo) and combining differential feature convolutional neural network (CDF-CNN). Similar to the tattoo attached on the skin, the electrode allows complying with the skin deformation comfortably and accommodation the local strains, providing long-term and robust monitoring of sEMG signals. Benefited from the convenient and low-cost fabrication and transferring to the skin surface, the large-area E-tattoo electrodes were applied to alleviate the requirement of controlling the accurate position of traditional Ag/AgCl electrodes. Moreover, a "sEMG feature map" is proposed by combining differential feature (CDF) to extract deep abstract features to improve the recognition effect, achieving 97.63% on average when using only two channels to classify 8 gestures. The proposed system is efficient, comfortable, natural and low-cost, which will help to facilitate the development and application of sEMG prosthesis.