A novel ECT image reconstruction method based on the deblurring diffusion model

Abstract

To address the challenge of low reconstruction accuracy in Electrical Capacitance Tomography (ECT) images, we propose an imaging method based on a deblurring diffusion model, transforming the image reconstruction into image generation. In cases where the noise type affecting the ECT image reconstruction cannot be determined, inspired by the cold diffusion model, we use an alpha kernel instead of Gaussian noise to perform the image blurring process. During the deblurring process, we utilize the U-Net as the backbone network and incorporated the Linear Attention module to improve imaging accuracy. Additionally, recognizing the similarity between image error metrics and the Dice function, we integrate the Dice function into the loss. Simulation experiments demonstrate that the image errors and correlation coefficients using the proposed method outperform those achieved with the LBP, Tikhonov, Landweber, and classic CNNs. Finally, we deploy the proposed algorithm to the actual ECT system.