Adversarial Resolution Enhancement for Electrical Capacitance Tomography Image Reconstruction

Abstract

High-quality image reconstruction is essential for many electrical capacitance tomography (CT) applications. Raw capacitance measurements are used in the literature to generate low-resolution images. However, such low-resolution images are not sufficient for proper functionality of most systems. In this paper, we propose a novel adversarial resolution enhancement (ARE-ECT) model to reconstruct high-resolution images of inner distributions based on low-quality initial images, which are generated from the capacitance measurements. The proposed model uses a UNet as the generator of a conditional generative adversarial network (CGAN). The generator’s input is set to the low-resolution image rather than the typical random input signal. Additionally, the CGAN is conditioned by the input low-resolution image itself. For evaluation purposes, a massive ECT dataset of 320 K synthetic image–measurement pairs was created. This dataset is used for training, validating, and testing the proposed model. New flow patterns, which are not exposed to the model during the training phase, are used to evaluate the feasibility and generalization ability of the ARE-ECT model. The superiority of ARE-ECT, in the efficient generation of more accurate ECT images than traditional and other deep learning-based image reconstruction algorithms, is proved by the evaluation results. The ARE-ECT model achieved an average image correlation coefficient of more than 98.8% and an average relative image error about 0.1%.

Generative-Adversarial-Network-Based Image Reconstruction for the Capacitively Coupled Electrical Impedance Tomography of Stroke

“…The training time was about 5.5 h. During training, the mean relative stand (8) was calculated on the testing part of the training dataset (Figure 13). To evaluate the network training quality, we adopted several simple pixe metrics such as the L2 norm (root mean square error, RMSE), the 2D co To evaluate the network training quality, we adopted several simple pixel-to-pixel metrics such as the L2 norm (root mean square error, RMSE), the 2D correlation coefficient, peak signal-noise ratio (PSNR), and structural similarity index (SSIM) [58,59] defined as follows:…”

Image Reconstruction Using Supervised Learning in Wearable Electrical Impedance Tomography of the Thorax

“…The latent vector is intended to provide sufficient intraclass variability, which helps to obtain a more reliable output. Such an approach works well in the case of a relatively small condition vector, for example, as in the case of 12 electrodes corresponding to the size 66 of the condition vector [70]. When the condition size increases, the network starts to lose stability, and some modifications are required.…”

“…Different architectures of ANNs were proposed for electrical tomography. These include feed-forward neural networks (FFNN) used in ECT [ 37 ], single-hidden layer feed-forward neural networks (SLFNs) used in EIT [ 35 ], Hopfield networks used in ECT [ 38 ], fully connected layers used in EIT [ 39 , 40 ], U-Net used both in ECT and EIT [ 33 , 41 ], a generative adversarial network used in both ECT [ 42 ] and EIT [ 43 ], and SegNet used in electrical resistance tomography (ERT) [ 44 ]. For capacitively coupled electrical resistance tomography, an approach based on convolutional neural networks (CNN) was proposed [ 45 ].…”

Review of Selected Advances in Electrical Capacitance Volume Tomography for Multiphase Flow Monitoring

“…The recent work by Deab et al [78] utilizes a conditional adversarial ML algorithm to improve the image resolution of ECT images. The proposed model consists of a UNet-based generator and a discriminator.…”

“…Recent advances in hardware and machine learning have ushered in new possibilities toward this objective. Some of the latest work published in the last few years [74,76,78] suggests improved image reconstruction performance provided by MLbased algorithms applied to ECT/ECVT. Additionally, a properly trained model can also image geometries that were previously considered almost unfeasible due to the nonlinear nature of the ECT/ECVT inverse problem.…”