V-Net Deep Imaging Method for Electrical Resistance Tomography

Monitoring the multiphase flow distribution has been a challenge in the industrial processes to improve the control efficiency and optimize the production. Electrical resistance tomography is a visualization technology that can be used to solve such a problem. However, image reconstruction of electrical resistance tomography is a nonlinear and ill-posed mathematical problem. To solve this problem, a supervised V-Net deep imaging method is proposed. A new 33-layer network based on convolutional neural network consists of three sequentially connected function modules, i.e. initial imaging module, feature extraction module, and image reconstruction module. Residual connection, and jump connection are used to increase the forward information flow and reverse gradient flow of V-Net. The loss function is composed of L
₂
regularization, cross entropy of layer 33, cross entropy of layer 5, which is used to constrain and monitor the imaging process. The presented network structure is capable of reconstructing the complex medium distribution with electrical resistance tomography. Experimental results show that the imaging quality of the V-Net imaging method is higher than the linear back projection algorithm, Tikhonov regularization algorithm and the latest related networks for image reconstruction of electrical resistance tomography.