Simultaneous Conductivity and Permeability Reconstructions for Electromagnetic Tomography Using Deep Learning

Electromagnetic tomography (EMT) is a research hotspot in electrical tomography, which has wide application prospect for multiphase flow measurement. The existing EMT usually visualizes the distributions of conductivity or permeability separately. In order to realize the simultaneous imaging of different electromagnetic characteristics in the measurement area and improve the quality of the reconstructed images, a deep learning based multi-parameter EMT method is proposed in this paper. Firstly, the information from the mutual inductance and magnetic induction intensity of the imaging area is measured respectively. Then, the Landweber algorithm is used to reconstruct the initial conductivity and permeability images using above measurements. Finally, the initial images are input into the improved DeepLabv3 network for image segmentation and the images of conductivity and permeability distributions with clear boundary and accurate size and position are output. The images reconstructed by the improved DeepLabv3 network are compared with those from traditional methods, UNet++, LinkNet and PAN networks through the simulation and experiment. The experimental results show that our method achieves RMSE of 0.1667, CC of 0.6984 and SSIM of 0.6542 on average for permeability distribution reconstruction, and RMSE of 0.1907, CC of 0.7791 and SSIM of 0.7538 on average for conductivity distribution reconstruction. These results prove that the proposed method can simultaneously obtain the conductivity and permeability distributions with high-quality reconstructed images. Our code is publicly available at https://github.com/Tougerr/Landweber-DLv3.