A convolutional neural network based electrical impedance tomography method for skin-like hydrogel sensing

The technique of electrical impedance tomography (EIT) offers significant benefits in designing large-area and flex-ible tactile sensors for robotic skin because it allows using only a small number of electrodes and simple structures without any internal wires. However, the poor-quality reconstruction images remain a great challenge for its application. In this paper, we presented a convolution neural network based EIT reconstruction algorithm (EIT-CNN) to improve the reconstruction quality of EIT-based tactile sensing. To verify the proposed algorithm, we conducted one-point numerical simulations. The simulation results showed that the proposed method can achieve better performance in effectively eliminating image artifacts and noise. Further, we performed a realistic multi-point experiment on a skin-like hydrogel tactile sensor. The results show the excellent generalization of the method and the potential for force strength detection.