Seismic exploration is an important tool for the detection of road diseases. However, since engineering seismic exploration usually deals with near-surface problems, its detection is complex and difficult. Time-frequency analysis is an important seismic attribute extraction method, which can provide hidden information that is difficult to obtain from seismic profiles, which can effectively help to identify subsurface structures and various types of disease. The S-transform is an important linear time-frequency analysis method, but the window function is fixed during its time-frequency feature extraction, resulting in a shift of the spectrum to higher frequencies, which reduces the accuracy of the time-frequency analysis. The unscaled S-transform, which removes the linear frequency term in the window function, overcomes the above problem to some extent, but affects the temporal resolution of the spectrum in the low-frequency region. To this end, we propose a modified frequency-domain unscaled S-transform method (MFUST) to perform the time-frequency decomposition of seismic signals, and the proposed method adds additional parameters to its window function, which ensures the time-frequency accuracy while realizing the improvement of the spectrum in terms of temporal resolution through the adjustment of the parameters. The effectiveness of the proposed method is verified using synthetic numerical experiments and a real data test.

Seismic inversion and petrophysical inversion are the most common methods used in exploration geophysics to obtain elastic and petrophysical parameters, which are essential for reservoir characterization. However, they are commonly ill-posed problems and both are usually performed independently. Recently, deep learning has been successfully applied to the solution of inverse problems (e.g., seismic inversion and petrophysical inversion) by using large amounts of labeled training data to establish a mapping relationship between the input and the target. On the one hand, the performance of deep learning-based inversion depends heavily on diversity of the training dataset. However, the number of wells in actual production is limited, which greatly limits the application of deep learning-based inversion methods. On the other hand, deep learning-based inversion methods usually calculate elastic and petrophysical parameters independently, which lacks clear physical meaning and leads to large computational errors. To overcome these problems, by considering the spatial variability of elastic and petrophysical parameters from well-log data, a large amount of realistic pseudo-well-log and post-stack seismic data are first generated based on geostatistics to obtain the diversity of data required for network training. Meanwhile, we propose a fusion neural network architecture to build a physically meaningful network to simultaneously estimate acoustic impedance and porosity within the reservoir conditioned to post-stack seismic and limited well data. The fusion neural network consists of two subnetworks based on the spatio-temporal neural network. The method is validated by the application of real data and compared to traditional inversion approaches.

Seismic inversion is one of the effective techniques to obtain elastic parameters for reservoir characterization. Deep learning is widely used in seismic inversion and has yielded many satisfactory results. The performance of the existing deep learning-based seismic inversion methods mainly depends on the network structure and a large number of effective training datasets. However, due to the limitation of expensive acquisition costs, it is difficult to obtain enough effective training datasets for network training in seismic surveys. To this end, we develop a double-dual network structure that incorporates both physics and model information to alleviate the dependence of deep learning methods on training data and even enables unsupervised learning and inversion. One of the dual networks is responsible for using the physical information to constrain the inversion results and ensure the physical validity of the predictions. The other dual network is responsible for using the priori information from the model domain to constrain the inversion results and improve the stability of the predictions. Ultimately, the two dual networks are coupled by a loss function to realize labeled/unlabeled network training and inversion applications. We then implement the method in a synthetic model as well as field data. The results are compared with traditional data-driven seismic inversion method and physics-guided data-driven seismic inversion method, and it is shown that the proposed method outperforms these two methods.

Geophysical exploration is important for road construction and maintenance. Before road construction, geophysical exploration is required to detect the geological structure to ensure the construction of the road; after the road is completed, geophysical exploration is still required to detect diseases in time. As a traditional geophysical exploration method, seismic exploration is important in road detection for its large detection depth and high resolution. The seismic attribute information obtained from seismic data can reflect many hidden information. As an important seismic attribute extraction method, time-frequency analysis method can simultaneously describe the energy density and intensity of seismic signals at different times and frequencies, which is of great significance to geological interpretation. However, traditional time-frequency analysis methods are low resolution and insufficient focusing. In this paper, on the basis of linear time-frequency analysis, the L1 norm constraint will be introduced, and the time-frequency analysis method will be implemented from the perspective of inversion, so as to reduce the influence of the multi-solution of the method and improve the method’s resolution and focusing. In this paper, two numerical simulation data and one real seismic data of road detection are employed to test the proposed new method.

Ground-penetrating radar (GPR) attenuation compensation is an indispensable part of radar data processing. However, traditional compensation methods based on physical mechanisms rely heavily on quality factor (Q) and are noise-sensitive, such as the inverse Q-filter method. Inspired by the non-linear mapping capabilities of deep learning (DL), a series of DL-based attenuation compensation methods have been proposed to circumvent the limitations of traditional attenuation compensation methods. Among the existing DL-based methods, convolutional neural networks (CNNs) are widely used. However, the electromagnetic wave propagation in the subsurface is characterized by dual dynamic changes in space and time. CNN only extracts local morphological features cannot ensure the integrity of the GPR data in the time dimension, which increases the instability of the attenuation compensation results. Therefore, we build a network - spatio-temporal neural network (STNN), for attenuating compensation of GPR data. The bidirectional long short-term memory (Bi-LSTM) layer is added to the CNN-based network to simultaneously capture both spatial and temporal characteristics of electromagnetic wave propagation, thus giving clear physical meaning to the DL-based attenuation compensation network. Numerical and real data experiments show that the attenuation compensation results of the proposed method have better continuity and stability than those of CNN-based and the stabilized inverse Q-filter method.

Seismic acoustic impedance bridges the gap between post-stack seismic reflection data and reservoir parameters such as lithology and porosity, and hence it plays an important role in stratigraphic interpretation. Due to sedimentation and propagation effects, both seismic records and impedance are a type of spatio-temporal data, i.e., there should be coupling between adjacent data points. However, most deep learning inversion methods only consider local shape information, ignore the time-series characteristics of the data, and are demanding on the training data, which makes inversion more difficult and leads to low inversion accuracy. For this reason, we develop a spatio-temporal neural network (STNN) to perform post-stack impedance inversion. The network consists mainly of a convolutional neural network (CNN) block and a recurrent neural network (RNN) block in series. Thus, the proposed method can take full advantage of CNN and RNN to capture the dynamics and correlations of seismic series in the spatio-temporal levels, yielding more continuous and stable results. We use an overthrust model example and an actual data case to test the performances of the STNN and demonstrate its advantages over traditional deep learning (i.e., CNN) based impedance inversion. Through a series of numerical experiments, we find that STNN produces more geologically reliable results, which not only ensure coupling relationships between adjacent points in the vertical direction, but also reveal well the stratigraphic variations in the lateral direction.

Seismic inversion is a method used to identify spatial structure and obtain physical properties of underground strata by processing seismic data. As a data-driven approach, deep learning (DL) is widely used in pre-stack three-parameter inversion to solve its non-linearity and ill-posed problems. However, traditional DL-based methods involve the construction of a separate network for each task and thus ignore the correlations between different tasks. Multi-task learning (MTL) aims to promote the effectiveness of each task with implicit information amplification by training multiple tasks simultaneously. However, information sharing in conventional MTL may cause negative effects on parallel tasks. To solve this problem, a novel multi-task learning method, called pertinent multi-gate mixture-of-experts (PMMOE), was proposed for pre-stack three-parameter inversion. PMMOE introduces mixture-of-experts (MOE) structure for multi-task learning and creatively divides experts into three special experts and a shared expert. In PMMOE, the input data of different experts are discrepant, enabling the retrieval of different features for different tasks. Experiments revealed that our proposed method has higher accuracy than other methods, and the inversion results of synthetic data and field data further demonstrate the effectiveness of our proposed method.

Seismic attenuation compensation is a widely used technique for enhancing the resolution of non-stationary seismic data. An accurate and reliable quality factor (Q) is a prerequisite for attenuation compensation and an important indicator of oil and gas. Q-factor estimation and attenuation compensation are ill-posed inverse problems, and they are poorly robust when noise is present. Moreover, the process is usually performed in steps, which leads to an accumulation of errors. Deep learning-based methods have gained great popularity because of their powerful ability to obtain exact solutions for inverse problems. However, most deep learning frameworks rely heavily on huge training datasets and lack clear physical meaning. For these reasons, we propose a fusion neural network architecture to build a physically meaningful network to simultaneously implement interval-Q estimation and seismic attenuation compensation. The fusion neural network consists of two sub-networks based on spatio-temporal neural network. And, the method ensures the coupling relationship between Q-factor and non-attenuated/attenuated seismic data by building a loss function that contain both Q loss term and seismic data loss term. The training dataset is then used to update the sub-networks simultaneously to obtain a fusion network that achieves both interval-Q estimation and seismic attenuation compensation. We demonstrate the effectiveness of the proposed simultaneous interval-Q estimation and seismic attenuation compensation algorithm by applying both synthetic and field data examples.

Prediction of elastic parameters (e.g., P-, S-wave velocity, and density) from observed seismic data is one of the most common means of reservoir characterization. Recently, deep learning (DL), as a data-driven approach, has been attracting increasing interest in seismic inversion. DL is proven to have the potential to learn complex systems and solve inverse problems efficiently. One of the most key components of DL is the training dataset, and an effective training dataset is a prerequisite for the success of DL-based methods. In seismic inversion, the training dataset needs to be artificially expanded due to the limited number of actual training data pairs. Traditional approaches of using the exact Zoeppritz equation (EZE) or its approximations for training dataset construction have limitations, principally, the single interface assumption and the neglect of wave propagation effects. Alternatively, the analytical solution of the 1D wave equation (i.e., reflectivity method-RM) can simulate the full-wave, including transmission losses and internal multiples, and can be executed in a target-oriented manner. Inspiration from this, we develop a data-driven elastic parameter prediction method based on waveform formulation. The method uses RM to construct training dataset, which both compensates for the inadequate training dataset in data-driven seismic inversion and improves the accuracy of the inversion results. We implement the method in a synthetic model as well as field data. The results are compared with model-driven methods (EZE and RM) and data-driven method based on EZE, and it is shown that the proposed method outperforms these three methods.

Prediction of lithology/fluid (LF) properties from seismic data can be very valuable in all phases of oil and gas exploration and production, but the resolution and accuracy of predicted results are reduced due to band-limited wavelet and noise of seismic data. Deep learning can review data, discover specific trends and patterns that would not be apparent to humans, and has been successfully used in many applications, including geophysics. Also, time-frequency (T-F) analysis tools can show how the energy of the signal is distributed over the 2-D T-F space, which helps to exploit the features produced by the concentration of signal energy. In this letter, we propose a novel hybrid approach for predicting LF properties, including oil-sand, brine-sand, and shale and evaluating their uncertainty, which aims at combining the benefits of T-F analysis method based on inverse spectral decomposition (ISD) and one-dimensional convolutional neural network (1D-CNN). The proposed method can provide more details about thinner layers and suppress noise to some extent using T-F spectrum obtained by ISD, and capture more relevant features from the input using 1D-CNN at different levels similar to a human brain, and thus, can significantly improve the resolution and accuracy of the predicted results. The proposed method was applied to a real 3-D post-stack seismic data and validated through a blind well test and comparison with the conventional methods.

Prediction of lithology/fluid (LF) characteristics is always the bottleneck problem and difficulty of reservoir characterization. Deep-learning-based data-driven methods can review data and find specific trends and patterns that would not be apparent to humans, and have been successfully used in many geophysical applications including LF prediction (LFP). However, the above methods mostly predict LF point-by-point, which means that the spatial correlation of LF is not considered. When the predicted LF results are combined to form a 2-D/3-D image, the resulting image will be noisy or even geologically unreliable. To overcome these issues, we proposed a spatially coupled data-driven (convolutional neural network, CNN) approach for LFP from the poststack seismic data and well observations. Here, the vertical couplings of the LF are modeled by a Markov chain (MC) prior and the lateral continuity of the LF is further defined by a Markov random field (MRF) prior. We also proposed to perform spectral decomposition via inversion strategies (ISD) to get a time-frequency (TF) spectrum as the input of CNN. ISD helps make full use of the information hidden in the frequency domain of the poststack seismic data. Well-logs and poststack seismic data are integrated in a consistent manner to obtain predictions of the LF classes with the associated uncertainty statements. The LFP results of the proposed approach are more laterally continuous and geologically reliable than the LFP results of the point-by-point. We determined the effectiveness of this methodology on a 2-D synthetic model and a 3-D field seismic data set.