Permittivity Reconstruction in Electrical Capacitance Tomography Based on Visual Representation of Deep Neural Network

Abstract

Electrical Capacitance Tomography (ECT) has been developed for many years and made great progresses. Successful applications of ECT depend on the accuracy and speed of image reconstruction. In this paper, we propose a new image reconstruction method based on deep neural network. The proposed neural network mainly consists of three parts, i.e. a forward problem network, an inverse problem network and a permittivity prediction network. Compared to previous reconstruction algorithms, the benefit of our method is that it can not only reconstruct the object shape, but also predict its permittivity value with the preset resolution which is desired in many multiphase flow applications. In experiment, 10000 frames of simulation data and additional measured capacitance data were used. The results show that the proposed method can reconstruct images more accurately than typical iterative methods with real permittivity values of objects predicted correctly, and reduce the computational cost to about 2.24 seconds per frame for an image resolution of 200*200.