A hybrid of iterative Gauss–Newton and one-dimensional convolutional neural network for high-resolution electrical impedance tomography

Abstract

We developed a processing method using benefits of both iterative Gauss–Newton (IGN) and a one-dimensional convolutional neural network (1D-CNN) for high-resolution electrical impedance tomography. The proposed method logically combines conductivity images reconstructed by different methods. The accuracies of the mathematical IGN method, 1D-CNN method, and the proposed method were compared. Utilizing the ideal potential data obtained through simulations, along with the experimental potential data derived from cement samples, we reconstruct the conductivity distribution. When utilizing the simulation data, the IGN method produces larger errors in the reconstructed images as the size of the foreign object decreases. The proposed method reconstructs the position and size more accurately than the IGN and 1D-CNN methods. When utilizing the experimental data, 1D-CNN and proposed methods were more accurate in terms of the position and size than the IGN method.