Image-to-Image Translation with Generative Adversarial Network for Electrical Resistance Tomography Reconstruction

TL;DR

This paper introduces a GAN-based method to improve electrical tomography image reconstruction, achieving higher resolution and detail compared to traditional algorithms, thus enhancing multiphase-flow monitoring capabilities.

Contribution

It presents a novel application of Pix2Pix GAN for electrical capacitance tomography, significantly improving image quality over conventional methods.

Findings

Enhanced SSIM, PSNR, and PMSE metrics

High-resolution images with sharp boundaries

Overcomes mesh discretization limitations

Abstract

Electrical tomography techniques have been widely employed for multiphase-flow monitoring owing to their non invasive nature, intrinsic safety, and low cost. Nevertheless, conventional reconstructions struggle to capture fine details, which hampers broader adoption. Motivated by recent advances in deep learning, this study introduces a Pix2Pix generative adversarial network (GAN) to enhance image reconstruction in electrical capacitance tomography (ECT). Comprehensive simulated and experimental databases were established and multiple baseline reconstruction algorithms were implemented. The proposed GAN demonstrably improves quantitative metrics such as SSIM, PSNR, and PMSE, while qualitatively producing high resolution images with sharp boundaries that are no longer constrained by mesh discretization.