The main challenge in the inversion of seismic data to predict the petrophysical properties of hydrocarbon-saturated rocks is that the physical relations that link the data to the model properties often are nonlinear and the solution of the inverse problem is generally not unique. As a possible alternative to traditional stochastic optimization methods, we develop a method to adopt machine-learning algorithms by estimating relations between data and unknown variables from a training data set with limited computational cost. We develop a probabilistic approach for seismic petrophysical inversion based on physics-informed neural network (PINN) with a reparameterization network. The novelty of our approach includes the definition of a PINN algorithm in a probabilistic setting, the use of an additional neural network (NN) for rock-physics model hyperparameter estimation, and the implementation of approximate Bayesian computation to quantify the model uncertainty. The reparameterization network allows us to include unknown model parameters, such as rock-physics model hyperparameters. Our method predicts the most likely model of petrophysical variables based on the input seismic data set and the training data set and provides a quantification of the uncertainty of the model. The method is scalable and can be adapted to various geophysical inverse problems. We test the inversion on a North Sea data set with poststack and prestack data to obtain the prediction of petrophysical properties. Compared with regular NNs, the predictions of our method indicate higher accuracy in the predicted results and allow us to quantify the posterior uncertainty.

Lithofacies classification is a fundamental step to perform depositional and reservoir characterizations in the subsurface. However, such a classification is often hindered by limited data availability and biased and time-consuming analysis. Recent work has demonstrated the potential of image-based supervised deep learning analysis, specifically convolutional neural networks (CNN), to optimize lithofacies classification and interpretation using core images. While most works have used transfer learning to overcome limited datasets and simultaneously yield a high-accuracy prediction. This method raises some serious concerns regarding how the CNN model learns and makes a prediction as the model was originally trained with entirely different datasets. Here, we proposed an alternative approach by adopting a vision transformer model, known asFaciesViT, to mitigate this issue and provide improved lithofacies prediction. We also experimented with various CNN architectures as the baseline models and two different datasets to compare and evaluate the performance of our proposed model. The experimental results show that the proposed models significantly outperform the established CNN architecture models for both datasets and in all cases, achieving an f1 score and weighted average in all tested metrics of 95%. For the first time, this study highlights the application of the Vision Transformer model to a geological dataset. Our findings show that the FaciesViT model has several advantages over conventional CNN models, including (i) no hyperparameter fine-tuning and exhaustive data augmentation required to match the accuracy of CNN models; (ii) it can work with limited datasets; and (iii) it can better generalize the classification to a new, unseen dataset. Our study shows that the application of the Vision transformer could further optimize image recognition and classification in the geosciences and mitigate some of the issues related to the generalizability and the explainability of deep learning models. Furthermore, the implementation of our proposedFaciesViTmodel has been shown to improve the overall performance and reproducibility of image-based core lithofacies classification which is significant for subsurface reservoir characterization in different basins worldwide.

Seismic inversion plays a very useful role in the detailed stratigraphic interpretation of migrated seismic volumes by enabling the estimation of reservoir properties over the complete volume. Traditional and machine learning-based seismic inversion workflows are limited to inverting each seismic trace independently of other traces to estimate impedance profiles, leading to lateral discontinuities in the presence of noise and large geologic variations in the seismic data. In addition, machine learning-based approaches suffer the problem of overfitting if there is only a small number of wells on which the model is trained. We have developed a two-pronged strategy to overcome these problems. We present a temporal convolutional network that models seismic traces temporally. We further inject the spatial context for each trace into its estimations of the impedance profile. To counter the problem of limited labeled data, we also present a joint learning scheme whereby multiple data sets are simultaneously used for training, sharing beneficial information among each of the sets. This results in improvement in the generalization performance on all data sets. We have developed a case study of acoustic impedance inversion using the open-source SEAM and Marmousi 2 data sets. Our evaluations show that our proposed approach is able to estimate impedance in the presence of noisy seismic data and a limited number of well logs with greater robustness and spatial consistency. We compare and contrast our approach to other learning-based seismic inversion methodologies in the literature. On SEAM, we are able to obtain an average mean squared error of 0.0476, the lowest among all other methodologies.

Worker safety in the industrial, construction, and energy sectors is the primary concern of governments, organizations, and project managers. Traditional monitoring systems are tedious, laborious, subjective, expensive, and inaccurate. Therefore, the development of automated and accurate systems for real-time monitoring and detecting personal protective equipment (PPE) compliance is urgently needed. This study provides valuable insights and contributions to the field of workers’ safety in the industrial, construction, and energy sectors. It aims to evaluate the potential contribution of providing a robust and accurate real-time PPE compliance monitoring system using state-of-the-art deep learning techniques with real-world videos collected from surveillance cameras. Furthermore, it presents a systematic framework for developing, evaluating, and selecting the most suitable model for real-world applications, starting from data collection and ending with system deployment and maintenance. To the best of our knowledge, the collected dataset in this work is the largest PPE’s dataset, with approximately 7.7 million labels representing different PPEs in 386,011 video frames collected from several surveillance cameras. The techniques adopted to generate different models and evaluated for PPE are CenterNet, Vision Transformer, and YOLOv7. Moreover, an optimized real-time inference engine is developed leveraging the high-computing power of GPUs in the system. The experimental results demonstrate that the models developed using YOLO are the most accurate and can be better generalized to detect PPE compliance in new working environments, achieving mean average precisions of 92.36% and 66.70% on Test Set I and Test Set II , respectively. In addition, they achieve real-time detection of PPE compliance.

For large-scale seismic exploration in areas that lack even basic infrastructure, wired geophones are impractical because of the huge effort involved and their high deployment and operating costs. A network of wireless geophones capable of recording and transmitting data could be an inexpensive solution. However, a typical seismic survey can generate hundreds of terabytes of raw seismic data per day. It takes a huge amount of energy to transmit this massive amount of data from geophones to the on-site data collection center, thus making the transformation from pre-wired to wireless geophones a significant challenge. To reduce data traffic to the data center without putting additional strain on the geophone, a standalone and lightweight compressive sensing (CS) method is proposed in this work. The method takes advantage of the inherent sparsity in the seismic data to enable the geophone to sense data in a compressed manner. This significantly reduces the amount of data that needs to be recorded/transmitted by the geophone, making it energy efficient. However, instead of employing conventional optimization-based CS reconstruction methods, we propose an efficient implementation of a deep convolutional neural network (DCNN). This network processes the compressed data received at the collection center without any a priori assumptions about the underlying seismic signal statistics, making it appropriate for a wide range of seismic data. The use of CS for energy-efficient sensing and transmission combined with powerful DCNN for reconstruction yields a system that could achieve signal-to-noise ratio (SNR) of around 30 dB with a compression gain of 16 on a field data set. Finally, when compared with other methods, the proposed approach demonstrates significant superiority in maximizing compression gain and reconstruction quality for both synthetic and real field data sets.