Swell noise is a common issue in streamer seismic data. This type of noise can significantly obscure useful signals and degrade the quality of subsequent seismic data processing. Traditional filtering methods struggle to effectively suppress strong noise, while deep learning-based denoising approaches using conventional convolutional neural networks (CNNs) often suffer from signal leakage due to pooling-based downsampling. To address these issues, we propose a Haar discrete wavelet transform (HDWT) downsampling-based attention U-Net (WAUNet) to effectively suppress swell noise while mitigating signal leakage. Furthermore, to provide a high-quality training dataset, we construct a multi-noise-level augmented dataset by combining real swell noise, clean synthetic data, and processed field data. Experimental results demonstrate that, compared to the self-supervised denoising approach and Attention-UNet, the proposed method achieves superior denoising performance on both synthetic and field seismic data.

Full-waveform inversion (FWI) is a critical geophysical imaging technique, renowned for its ability to generate high-precision subsurface structural models. However, FWI is a highly nonlinear and ill-posed inverse problem, necessitating appropriate regularization to incorporate prior information. Recently, the plug-and-play (PnP) method has shown promise in addressing various inverse problems by leveraging pretrained deep learning (DL)-based denoisers as priors, thus eliminating the need for explicit regularization functions. Building on this approach, we employ a pretrained DRUNet denoiser from the field of computer vision to implement regularization constraints for FWI via the PnP method. We propose a novel FWI framework (FWISD-DRU) that integrates graph space Sinkhorn distance (GS-SD) with the DRUNet denoiser to enhance FWI quality. By utilizing a bias-free denoising network architecture and additional noise-level map inputs, our approach improves the adaptability of the pretrained network to various geological images. Numerical tests on typical geological models validate the universality and effectiveness of our method. The experimental results demonstrate that our approach, incorporating the DRUNet denoiser, produces more accurate and higher resolution inversion results than total variation (TV) regularization and the denoising convolutional neural network (DnCNN) and FFDNet denoisers. This advantage is particularly pronounced under challenging conditions such as poor initial models, noisy observed data, and missing low-frequency components.

Seismic inversion methods for PP waves in transversely isotropic media with a vertical axis of symmetry (VTI) have been extensively studied. In comparison, seismic shear waves (SV-SV and SH-SH waves) exhibit much higher sensitivity to anisotropy. Among them, the SV-SV wave in VTI media displays nonelliptical anisotropy, which is more complex than that of the SH-SH wave. Incorporating anisotropic effects in the inversion of SV-SV wave data is, therefore, crucial for accurate subsurface characterization. In this study, we propose a seismic amplitude inversion method for SV-SV waves in VTI media based on a modified approximate equation for the SV-wave reflection coefficient. This method facilitates the inversion of three key parameters: (A) vertical shear wave impedance, (B) shear modulus related to the anellipticity anisotropy parameter, and (C) vertical shear wave velocity. Additionally, the effective parameter$\sigma $, which governs the influence of anisotropy on the SV-wave, can be derived from these parameters. Synthetic tests demonstrate that the proposed inversion method maintains strong robustness against varying noise levels. Application to field data from a shale oil exploration area validates the method’s efficacy in recovering both elastic parameters and the anisotropic characteristics of fine layers. This work provides a reliable tool for the characterization of anisotropic reservoir formations.

Seismic coherent noise is often found in post-stack seismic data, which contaminates the resolution and integrity of seismic images. It is difficult to remove the coherent noise since the features of coherent noise, e.g., frequency, are highly related to signals. Recently, deep learning has proven to be uniquely advantageous in image denoise problems. To enhance the quality of the post-stack seismic image, in this letter, we propose a novel deep-residual-learning-based neural network named DR-Unet to efficiently learn the features of seismic coherent noise. It includes an encoder branch and a decoder branch. Moreover, in order to collect enough training data, we propose a workflow that adds real seismic noise into synthetic seismic data to construct the training data. Experiments show that the proposed method can achieve good denoising results in both synthetic and field seismic data, even better than the traditional method.

Shear waves have been used for oil and gas exploration for decades. Reflected seismic shear waves (SV-SV and SH-SH) have shown good capability in gas-cloud imaging and geological interpretation. Compared with the longitudinal wave, shear waves are more sensitive to anisotropy. The SV-SV wave traveltime may significantly deviates from the hyperbolic form in anisotropic media. Conventional hyperbolic normal moveout (NMO) correction is inadequate for handling the far-offset SV-SV wave data and is insufficient to recover the true model parameters. VTI (transverse isotropy with a vertical axis of symmetry) is one of the most common types of anisotropy in sedimentary basins. Nonhyperbolic analysis in VTI media is important, but the influence of VTI on the SV-SV wave kinematics is still ambiguous. To solve this problem, we first analyze the influence of varied anisotropy parameters and velocity ratio (the ratio of the vertical P-wave velocity to the vertical S-wave velocity) on SV-SV wave traveltime in VTI media. Secondly, we evaluate three SV-SV waves approximate traveltime equations in terms of their accuracy with parameters and offset-depth ratio (x/z). Finally, a nonhyperbolic normal moveout (NMO) correction method for SV-SV wave in VTI media is established based on the two-parameter (normal moveout velocityVS2and effective anisotropic parameter ζ) approximate equation, in which both parameters are inverted from the reflection events. It is applied to a seismic shear wave dataset which shows obvious VTI characteristics acquired from Qaidam Basin in China. The reflection events in common-middle-point gather show typical “hockey stick” phenomenon after a conventional hyperbolic NMO correction, while the proposed method can flatten reflection events at full offset, with the NMO velocityVS2inverted from the near-offset reflection events, and the effective anisotropic parameter ζ inverted from the far-offset reflection events. The following stacked seismic imaging by using the proposed method also shows great improvement. Besides, more far-offset information can retain, and this has a significant impact on AVO analysis and interpretation of shear wave seismic data.

Fault detection is an essential component of seismic interpretation and plays a crucial role in industrial processes. However, it is also one of the main challenges, especially in delineating faults in 3D seismic data. Recently, the rapidly developing technology, deep learning, has proven to be a powerful tool for this task. A number of neural networks have been proposed for this purpose by regarding 3D fault detection as a semantic segmentation task. To further enhance the effectiveness of the deep learning methods, we propose a novel network architecture, named R2SE-Unet, to solve the 3D segmentation problem. In the neural network, we design a recurrent residual-SE convolution unit (RRCU-SE) that integrates the residual learning and Squeeze-Excitation module to store information in 3D seismic data. This component promotes the spread of 3D volumetric information and aids in learning spatial dependencies in 3D images. In addition, to reduce the impact of insufficient spatial resolution resulting from the base architecture of U-net, we add an attention unit between skip connection operations. These two new units enable our R2SE-Unet to exploit semantic information more accurately in the feature maps. After many experiments on region-based loss functions and distribution-based loss functions, we also propose a novel loss function, which takes the advantage of generalized dice (GDice) loss and balanced binary cross entropy (b-BCE) loss, named Gdice-bce, to effectively train R2SE-Unet. Although only synthetic seismic data samples are used to train the network parameters, our R2SE-Unet could produce more reliable fault feature maps on field seismic data than two other conventional fault detection neural networks. Thus, the proposed neural network is easy to train and reliably works for seismic fault interpretation on field seismic data.