A transformer-based model for two-phase flow imaging by electrical resistance tomography

Electrical resistance tomography (ERT) is a powerful method for two-phase flow measurement in various applications, however, the image reconstruction quality of ERT is often unsatisfactory due to the ill-posed nature of the inverse problem. In order to improve the imaging quality of ERT, a deep learning approach based on Transformer architecture is proposed. To reduce the risk of overfitting on small datasets caused by the weak inductive bias, a convolutional layer is used before the Transformer Encoder module in this paper. Trained on boundary voltages and images of the target location, the Transformer model establishes a nonlinear mapping relationship between them. The simulation results demonstrate that the proposed Transformer model achieves superior image reconstruction performance compared to the traditional image reconstruction algorithm, and have good generalization and noise resistance capabilities.