Deep neural networks have received considerable attention in clinical imaging, particularly with respect to the reduction of radiation risk. Lowering the radiation dose by reducing the photon flux inevitably results in the degradation of the scanned image quality. Thus, researchers have sought to exploit deep convolutional neural networks (DCNNs) to map low-quality, low-dose images to higher-dose, higher-quality images, thereby minimizing the associated radiation hazard. Conversely, computed tomography (CT) measurements of geomaterials are not limited by the radiation dose. In contrast to the human body, however, geomaterials may be comprised of high-density constituents causing increased attenuation of the X-rays. Consequently, higher-dose images are required to obtain an acceptable scan quality. The problem of prolonged acquisition times is particularly severe for micro-CT based scanning technologies. Depending on the sample size and exposure time settings, a single scan may require several hours to complete. This is of particular concern if phenomena with an exponential temperature dependency are to be elucidated. A process may happen too fast to be adequately captured by CT scanning. To address the aforementioned issues, we apply DCNNs to improve the quality of rock CT images and reduce exposure times by more than 60%, simultaneously. We highlight current results based on micro-CT derived datasets and apply transfer learning to improve DCNN results without increasing training time. The approach is applicable to any computed tomography technology. Furthermore, we contrast the performance of the DCNN trained by minimizing different loss functions such as mean squared error and structural similarity index.
Computed tomography (CT) is an important tool to characterize rock samples allowing quantification of physical properties in 3D and 4D. The accuracy of a property delineated from CT data is strongly correlated with the CT image quality. In general, high-quality, lower noise CT Images mandate greater exposure times. With increasing exposure time, however, more wear is put on the X-Ray tube and longer cooldown periods are required, inevitably limiting the temporal resolution of the particular phenomena under investigation. In this work, we propose a deep convolutional neural network (DCNN) based approach to improve the quality of images collected during reduced exposure time scans. First, we convolve long exposure time images from medical CT scanner with a blur kernel to mimic the degradation caused because of reduced exposure time scanning. Subsequently, utilizing the high- and low-quality scan stacks, we train a DCNN. The trained network enables us to restore any low-quality scan for which high-quality reference is not available. Furthermore, we investigate several factors affecting the DCNN performance such as the number of training images, transfer learning strategies, and loss functions. The results indicate that the number of training images is an important factor since the predictive capability of the DCNN improves as the number of training images increases. We illustrate, however, that the requirement for a large training dataset can be reduced by exploiting transfer learning. In addition, training the DCNN on mean squared error (MSE) as a loss function outperforms both mean absolute error (MAE) and Peak signal-to-noise ratio (PSNR) loss functions with respect to image quality metrics. The presented approach enables the prediction of high-quality images from low exposure CT images. Consequently, this allows for continued scanning without the need for X-Ray tube to cool down, thereby maximizing the temporal resolution. This is of particular value for any core flood experiment seeking to capture the underlying dynamics.
Gradient based traveltime tomography, which aims to minimize the difference between modeled and observed first arrival times, is a highly non-linear optimization problem. Stabilization of this inverse problem often requires employing regularization. While regularization helps avoid local minima solutions, it might cause low resolution tomograms because of its inherent smoothing property. On the other hand, although conventional ray-based tomography can be robust in terms of the uniqueness of the solution, it suffers from the limitations inherent in ray tracing, which limits its use in complex media. To mitigate the aforementioned drawbacks of gradient and ray-based tomography, we approach the problem in a completely novel way leveraging data-driven inversion techniques based on training deep convolutional neural networks (DCNN). Since DCNN often face challenges in detecting high level features from the relatively smooth traveltime data, we use this type of network to map horizontal changes in observed first arrival traveltimes caused by a source shift to lateral velocity variations. The relationship between them is explained by a linearized eikonal equation. Construction of the velocity models from this predicted lateral variation requires information from, for example, a vertical well-log in the area. This vertical profile is then used to build a tomogram from the output of the network. Both synthetic and field data results verify that the suggested approach estimates the velocity models reliably. Because of the limited depth penetration of first arrival traveltimes, the method is particularly favorable for near-surface applications.
In Geological Carbon Sequestration (GCS), mineralization is a secure carbon dioxide (CO2) trapping mechanism to prevent possible leakage at later stage of the GCS project. Modeling of the mineralization during GCS relies on numerical reservoir simulation, but the computational cost is prohibitively high due to the complex physical processes. Therefore, deep learning (DL) models can be used as a computationally cheaper and at the same time, reliable alternative to the conventional numerical simulators. In this work, we have developed a DL approach to effectively predict the dissolution and precipitation of various important minerals, including Anorthite, Kaolinite, and Calcite during CO2 injection into deep saline aquifers. We established a reservoir model to simulate the process of geological CO2 storage. About 750 simulations were performed in order to generate a comprehensive dataset for training DL models. Fourier Neural Operator (FNO) models were trained on the simulated dataset, which take the reservoir and well properties along with time information as input and predict the precipitation and dissolution of minerals in space and time scales. During the training process, root-mean-squared-error (RMSE) was chosen as the loss function to avoid overfitting. To gauge prediction performance, we applied the trained model to predict the concentrations of different mineral on the test dataset, which is 10% of the entire dataset, and two metrics, including the average absolute percentage error (AAPE) and the coefficient of determination (R2) were adopted. The R2 value was found to be around 0.95 for calcite model, 0.94 for Kaolinite model, and 0.93 for Anorthite model. The R2 was calculated for all trainable points from the predictions and ground truth. On the other hand, the average AAPE for all the mappings was calculated around 1%, which demonstrates that the trained model can effectively predict the temporal and spatial evolution of the mineral concentrations. The prediction CPU time (0.2 seconds/case) by the model is much lower than that of the physics-based reservoir simulator (3600 seconds/case). Therefore, the proposed method offers predictions as accurate as our physics-based reservoir simulations, while provides a huge saving of computation time. To the authors' best knowledge, prediction of the precipitation and dissolution of minerals in a supervised learning approach using the simulation data has not been studied before in the literature. The DL models developed in this study can serve as a computationally faster alternative to conventional numerical simulators to assess mineralization trapping in GCS projects especially for the mineral trapping mechanism.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.