Image Reconstruction for Electrostatic Tomography Based on Residual Network Considering the Prior Knowledge of Boundary Measurement

Electrostatic tomography (EST) is a passive tomographic image method, which determines that the number of independent measurements and the signal-to-noise ratio (SNR) of its signal are smaller than those of active excited electrical tomography (ET) such as electrical capacitance tomography (ECT). The nonlinearity and ill-posed property of EST are more prominent due to the nature of passive measurement. The traditional iterative approximation method is insufficient to express the nonlinear nature of image reconstruction, so the accuracy of the reconstructed image is low. To solve this problem, a residual network (ResNet) model is proposed in this paper. A new loss function based on the prior knowledge of the boundary measurement of EST is proposed to make the model fit the imaging principle better. In order to build the dataset, 3500 samples are generated through simulation and divided into training set and validation set. Several typical flow patterns are simulated separately as test set. The reconstructed results of the model on the test set have high accuracy compared with some conventional algorithms. When Gaussian white noise with SNR of 10 dB, 15 dB and 20dB is added to the test set, the reconstructed results of the model can still represent the approximate position of the sample, which proves the robustness of the model.