Electrical Resistance Tomography Image Reconstruction Based on Res2net4 Network

Electrical resistance tomography (ERT) is a non-invasive process measurement technology, which has the advantages of high speed, low cost and strong robustness. It has a broad application prospect in multiphase flow monitoring and measurement. Image reconstruction is the key part of electrical resistance tomography. Because it is a nonlinear and ill posed mathematical problem, the traditional image reconstruction methods cannot achieve high accuracy. Therefore, this paper proposes a deep learning method based on Res2net4 network for ERT image reconstruction. The network is composed of two large residual modules, each of which is grouped and then connected by residual within the group to fuse more feature scales. The network loss function is composed of cross entropy and L2 regularization term in the last layer, which is used to constrain and monitor the imaging process. The simulation and experimental results show that the proposed network can reconstruct complex medium distribution with a higher imaging quality than LBP, Landweber, Tikhonov and other traditional image reconstruction algorithms.