Logging-while-drilling (LWD) azimuthal imaging is a widely used well-logging technique in modern geological resource exploration. However, due to the measurement principles and data transmission capacity, the circumferential resolution of current techniques is very limited. In this article, we propose a deep convolutional network-based algorithm called azimuthal image super-resolution (AzSR), which is capable of reconstructing high-resolution borehole images with 128 fans from noisy azimuthal responses of 4/8/16 fans. To make the proposed algorithm more suitable for AzSR, techniques such as sample synthesis, circular padding, and special loss terms are introduced. The advantages and effectiveness of the proposed AzSR algorithm are demonstrated through systematic experiments and real-world applications. The results show that the proposed AzSR has significant advantages over existing algorithms in terms of noise robustness, detail reconstruction, and resolution improvement. With the super-resolution results of AzSR, detailed information about lithological interfaces, local structure, and thin layers can be clearly revealed. This will be of great value for decision-making during geosteering drilling and for fine-scale geological interpretation after drilling.

Ground-roll is a typical Rayleigh-type interference noise in field seismic data, which is characterized by low frequency, low velocity and high amplitude. Since it will interfere effective seismic signals and severely degrade the signal-to-noise ratio of observed seismic records, many approaches have been developed for ground-roll attenuation or separation. In this letter, we proposed an improved ground-roll separation algorithm through the combination of deep learning based low-frequency generation and dictionary learning based low-frequency reconstruction. Moreover, to utilize the inter-band morphological similarity prior in seismic response, we introduce the morphological similarity constraint into the learning approach of pseudo low-frequency generation networks. Experiments demonstrate that compared to previous methods, the introduced the morphological similarity loss can effectively improve the quality of generated pseudo low-frequency signals, which results in better low-frequency reflection reconstruction and ground-roll separation performances.

Seismic stratigraphic interpretation plays an important role in geophysics and geosciences. Recently, deep learning has been explored for seismic stratigraphic interpretation. However, deep learning-based interpretation methods usually require sufficient labeled samples. This is often too hard to be satisfied in field seismic interpretation. In this paper, we propose a deep active learning-based method to address this issue. Active learning typically exploits prediction uncertainty to reduce labeling effort. We found that uncertainty of prediction is easily obtained in the field of seismic interpretation. Since adjacent seismic images are very similar, they should have similar predictions. When the model performs poorly, the predictions of adjacent images will differ significantly. Thus, the uncertainty can be easily obtained by measuring the similarity of the predictions of adjacent seismic images. Then, data with the highest uncertainty is annotated by geological expert and used for the next round of training. For few-shot active learning, initial models obtained by different initial training sets are quite different. We combine deep clustering and uncertainty sampling to select initial training datasets, with which a good initial model can be obtained. To improve generalization, we introduce a random thin plate spine transformation to simulate changes of terrain. We apply the proposed method to the F3 field seismic data. The results demonstrated that the proposed method can effectively improve performance of learned seismic interpretation network with very limited labeled samples.

Accurate estimation of volumetric seismic dip is of great significance for subsequent seismic processing and interpretation works. Recently, with the development of deep learning techniques, convolutional networks are also applied for seismic dip estimation. Compared with traditional approaches, estimating dips with convolutional networks is not only more efficient but also shows great promise in accuracy and robustness. However, if we take dips estimated by traditional approaches as labels and train networks on the field seismic data directly, the accuracy and robustness of learned networks are influenced due to the error in dip labels. An alternative solution is synthesizing realistic seismic samples with accurate dip labels. However, we find that due to the differences in seismic responses and structural patterns between the synthetic and field seismic data, networks directly learned from synthetic samples cannot guarantee their generalization on the field seismic data. To overcome these drawbacks, we develop a transfer learning approach for improvement. The proposed approach pretrains the dip estimation network on synthetic seismic samples at first and then transfers it to the targeted field seismic data with a domain knowledge-inspired fine-tuning process. Moreover, the proposed approach also highlights the combination of deep learning techniques and domain knowledge in seismic processing—several subtle realizations, such as knowledge-driven sample augmentation, knowledge constrained loss function, and knowledge motivated transfer learning strategy, are introduced, which greatly enhance the learning of the seismic dip estimation network. The advantages of the proposed approach in accuracy, robustness, and resolution are validated by applying the estimated dips for structural filtering and curvature extraction on the Netherlands F3 and Kerry3D seismic data, which further confirms its practicality in the real-world application. We believe that the proposed approach has provided an effective improved way for further seismic dip estimation practices, and the present domain knowledge constrained deep learning case will also inspire researchers in the same discipline.

Recently, a multitask learning framework named M: multitask, R: global residual skip connection structure, U: encoder–decoder structure of U-Net, D: dense skip connection structure, and SR: super-resolution (M-RUDSR) has successfully improved the accuracy of full-waveform inversion (FWI) results by enhancing the resolution of the seismic velocity model. However, M-RUDSR does not make full use of seismic data even though it contains high wavenumber information, which can help enhance the resolution of the velocity model. Moreover, the effects of employing seismic data realized by simply increasing the model’s input and output channels are limited since the seismic velocity model and seismic data are in different frequency bands. Therefore, we propose to consider super-resolution (SR) of seismic data and its edge images as supplementary auxiliary tasks of the seismic velocity model SR. Besides, the proposed method named M-RUDSRv2 improves the resolution of the seismic velocity model leveraging a three-step learning strategy. First, the model in M-RUDSRv2 is trained preliminarily on the specific data where the seismic velocity model and seismic data are in the same blurring levels. Then, the pretrained model is fine-tuned on the extensive data, where the seismic velocity model and seismic data are in various kinds of blurring levels, to achieve strong generalization ability. Finally, the fitted model focuses on improving the resolution of the seismic velocity model by adjusting the parameters in the loss function. Comparative experiments on synthetic and field data validate the superior performance of M-RUDSRv2 compared with M-RUDSR in SR of the seismic velocity model.

Seismic inversion is aimed at building a mapping from low-resolution seismic data to high-resolution impedance data. Most of the traditional methods have satisfactory interpretability, and most parameters tend to have specific physical definitions. On the other hand, deep learning-based methods present poor interpretability as their prediction performance is not always clearly explainable. One of the significant challenges of the deep learning-based methods is to quantify the uncertainty of the model. The uncertainty includes aleatoric uncertainty and epistemic uncertainty, and epistemic uncertainty can be used to evaluate the predicted accuracy of the trained model. In this paper, we propose a new deep learning model called uncertainty backpropagation network (UB-Net) to perform impedance inversion. The proposed UB-Net is based on a closed-loop framework and can predict the impedance and the epistemic uncertainty simultaneously. UB-Net has three closed-loop data flows, whereby the predicted uncertainty is utilized as the weight of loss functions to improve the inversion accuracy. Experimental analyses demonstrate that UB-Net presents advanced inversion accuracy on both synthetic and real examples. Specifically, the mean absolute error (MAE) on synthetic examples drops by 40%, and the Pearson correlation coefficient (PCC) on real examples increases by 2%. Besides, compared with existing approaches, UB-Net presents superior spatial continuity and preserves more geological structures such as little faults in real examples.

Deep learning has been applied to tackle the seismic inversion problem, bringing more efficiency and accuracy. However, bad spatial continuity and poor generalizability limit the practical application. To solve these problems, we propose a 2D end-to-end seismic inversion method based on domain adaption. Firstly, the proposed 2D network learns the inversion mapping of seismic data under the constraint of domain adaption layer, which can reduce the difference between the features of real seismic data and synthetic seismic data, improving the generalization ability on real seismic data. Then, the trained model is finetuned with well logging data. In the first process, the spatial continuity of the inversion result is guaranteed by the 2D training scheme. Meanwhile, due to the constraint of the domain adaption layer, our model not only performs well on the synthetic data but also has good generalization ability on the real seismic data. And we carefully discuss the mechanism of domain adaption layer. In the second process, finetuning introduces well logging information, which can further improve the ability to invert details. Moreover, in order to improve the inversion accuracy on real seismic data, we develop a new training data generation method that can generate the synthetic samples close to the real samples, and a 2.5D training strategy is adopted to improve the continuity of the 3D data. The experiments on both synthetic and real seismic data show that our method performs better than both the recursive inversion method and the 1D closed-loop CNN methods.

Transverse relaxation T2 spectrum obtained by nuclear magnetic resonance (NMR) logging tools is an intuitive reflection of the pore size distribution for subsurface formation, which is valuable for petroleum reservoir characterization. However, the deployment of NMR logging tools is constrained by financial and operational factors, while NMR data are only available in very limited wells. This seriously limits its application in practices. Therefore, researchers try to synthesize NMR T2 spectra from more widely measured conventional logging data with the help of machine learning technologies. In the article, we propose the spectrum regression forest (SRF) algorithm for the prediction of NMR T2 spectra from conventional logging responses. Based on the experiment on the real-world well data of carbonate reservoir, the proposed algorithm is proved to provide effective NMR T2 spectrum predictions with accuracy amplitudes and consist morphology, which is believed to enhance the understanding of formation pore structures for future reservoir characterization practices.

Structural curvatures are widely used seismic attributes that help interpreters to understand both structural and stratigraphic features. Traditional structural curvature extractions are mainly calculated from dip estimations through lateral scanning of seismic events, which is not only a very time-costing approach but also influenced by parameter settings, seismic frequency, and data quality. In this article, we propose a deep learning-based volumetric curvature extraction approach that directly derives structural curvature volumes from the seismic response. To realize the above approach, we develop a suite of sample generation and augmentation methods to synthesize seismic samples with accurate curvature labels. Then, a multitask end-to-end convolutional neural network architecture and a geometric loss function are proposed to establish the volume mapping model from complex seismic responses to the most positive and negative curvature volumes. The performance of the proposed curvature extraction approach is evaluated on both the synthetic data and the Netherlands F3 field seismic data. Extensive experiments demonstrate that curvature volumes extracted with the proposed approach are not only more accurate and less influenced by the noises of poststack seismic data but also more friendly for structure interpretation. Therefore, we believe that our proposed deep learning curvature extraction approach can be a useful tool for further seismic structure interpretation practices.

Full waveform inversion (FWI) is a powerful tool for estimating the underground velocity model. However, it is computationally expensive and the resulting models tend to be not accurate enough. Thus, to improve the efficiency and accuracy of FWI, we propose a super-resolution (SR) method based on deep learning to enhance the resolution of the seismic velocity model. Since the edge images of the seismic velocity model are also widely used in geophysics, a multitask learning (MTL) network with hard parameter sharing is applied to perform the SR of the seismic velocity model and its edge images. The proposed MTL model dubbed M-RUDSR includes a global residual skip connection, an encoder-decoder structure of U-Net, and a dense skip connection structure. Besides, two networks for comparison, namely, RUDSR and M-RUSR, are proffered. RUDSR is a single-task version of M-RUDSR, whereas M-RUSR is a simplified version of M-RUDSR without a dense skip connection structure. Compared with RUDSR and M-RUSR, M-RUDSR produced the best results for all kinds of blurring levels and achieved better visual details. We found that FWI followed by SR can help reduce the computational cost of FWI in the high-frequency part of the spectrum, as well as achieve better high-frequency details recovery. The experimental results show that M-RUDSR is a practical recovery scheme in SR of the seismic velocity model and can be applied to a real data set efficiently.

Spatial variable estimation is a basic application of geostatistics. In general, this task is performed based on observations of limited points. For some cases, intensive observed data obtained from other sources are also available as the auxiliary variables. To utilize the auxiliary information in these data, methods such as regression kriging (RK) or cokriging are proposed. However, these methods all assume that the auxiliary variables keep linear correlation with the target variable implicitly, which is not satisfied in most cases. In this letter, through the combination of nonlinear machine learning mapping (MLM), we propose a novel hybrid method to relax the linear assumption of RK. The proposed method is applied to a real-world subsurface shale volume estimation task for demonstration. Compared with existing methods such as ordinary kriging, RK, and MLM, the relative estimation error reduction of the proposed method is larger than 10%. Meanwhile, the estimation resolution is also improved. This indicates that the proposed method provides an alternative way for further spatial variable estimation practices.

Borehole lithology discrimination is the foundation for formation evaluation and reservoir characterization. Due to the limitation of costing or accuracy, direct discrimination methods, such as borehole core and drilling cutting analysis, are unable to widely apply, while logging lithology interpretation provides an alternative solution for this task. To mitigate the influence of subjective bias, several machine learning algorithms, such as neural network, support vector machine, decision tree, and random forest (RF), have already been applied for logging lithology interpretation. However, the vast majority of preceding studies are simple applications that directly apply classification algorithms to the raw input space formed by logging curve values, only limited studies involved feature extraction or learning space transformation. In this letter, we propose a hybrid algorithm that combines the mean-shift algorithm and the RF algorithm for borehole lithology discrimination in the prototype similarity space. Experiments on data collected from nine different areas demonstrate that the proposed algorithm has significant advantages in accuracy compared with other algorithms, which provides a considerable alternative way for further machine learning-assisted logging lithology interpretation.