Seismic impedance inversion is essential to characterize hydrocarbon reservoir and detect fluids in field of geophysics. However, it is nonlinear and ill-posed due to unknown seismic wavelet, observed data band limitation and noise, but it also requires a forward operator that characterizes physical relation between measured data and model parameters. Deep learning methods have been successfully applied to solve geophysical inversion problems recently. It can obtain results with higher resolution compared to traditional inversion methods, but its performance often not fully explored for the lack of adequate labeled data (i.e., well logs) in training process. To alleviate this problem, we propose a semi-supervised learning workflow based on generative adversarial network (GAN) for acoustic impedance inversion. The workflow contains three networks: a generator, a discriminator and a forward model. The training of the generator and discriminator are guided by well logs and constrained by unlabeled data via the forward model. The benchmark models Marmousi2, SEAM and a field data are used to demonstrate the performance of our method. Results show that impedance predicted by the presented method, due to making use of both labeled and unlabeled data, are better consistent with ground truth than that of conventional deep learning methods.

In recent years, significant progress has been made in self-supervised seismic denoising. However, most of these methods focus on random noise attenuation, which is of little practical use for distributed acoustic sensing vertical seismic profile (DAS VSP) data with a large amount of spatially correlated noise. In this paper, we propose a new perspective to solve this problem, which is to redesign the blind spot training scheme by analyzing different texture features of effective signal and spatially correlated noise. Specifically, we fully analyze the correlation of different correlated noises in vertical and horizontal scales, and then use a novel blind masking strategy to carefully design the receptive field to expand the blind spot network (BSN) to neighborhood-mask network (NMN), which makes BSN still meet the assumption of noise pixel independence in the presence of a large amount of correlated noise. At the same time, we introduce a texture total variation (TTV) regularization to further eliminate correlated noise and preserve the local smooth structure of effective signal. In order to maximize the texture difference between signal and noise, we propose asymmetric-dilated convolution (ADConv), which has a large receptive field in specific direction and acts as an information filter in the network. Our method, also called SelfTexture, only uses observed noisy DAS VSP data to remove correlated noise in a self-supervised manner and further demonstrates the great potential of BSN in dealing with correlated noise. Extensive experiments on synthetic and field DAS VSP data validate the superior performance of our SelfTexture.

Seismic data processing plays a key role in the field of geophysics. The collected seismic data are inevitably contaminated by various types of noise, which makes the effective signals difficult to be accurately discriminated. A fundamental issue is how to improve the signal-to-noise ratio of seismic data. Due to the complex characteristics of noise and signals, it is a challenge for the denoising model to suppress noise and recover weak signals. To suppress random noise in seismic data, we propose a multi-scale deformable convolution neural network denoising model based on U-Net, named MSDC-Unet. The MSDC-Unet mainly contains modules of deformable convolution and dilated convolution. The deformable convolution can change the shape of the convolution kernel to adjust the shape of seismic signals to fit different features, while the dilated convolution with different dilation rates is used to extract feature information at different scales. Furthermore, we combine Charbonnier loss and structure similarity index measure (SSIM) to better characterize geological structures of seismic data. Several examples of synthetic and field seismic data demonstrate that the proposed method is effective in the comprehensive results in terms of quantitative metrics and visual effect of denoising, compared with two traditional denoising methods and two deep convolutional neural network denoising models.

Estimating the elastic parameters from prestack inversion is of great importance for reservoir characterization. Conventional three-parameter prestack inversion methods rely heavily on well logs, and it is difficult to obtain reliable inversion results in situations with limited numbers of wells. Alternatively, we have developed a joint inversion strategy, integrating the advantages of post- and prestack inversion, to deal with the situation. First, due to the high signal-to-noise ratio of poststack seismic data, the high-precision acoustic impedance (AI) inversion is conducted. Second, the exact Zoeppritz equation is used to establish the objective function of the prestack inversion. To better constrain the elastic parameters (P- and S-wave velocities and density), a new similarity measurement criterion, the gradient structure similarity (GSS), is defined to describe the structural similarity between the prestack inversion results and the inverted AI from the poststack inversion. Third, the LM optimization algorithm is used to solve the nonlinear objective function. Through the model test, we verify the effectiveness of our GSS regularization scheme. Some synthetic and field examples find that our method can provide more stable and accurate inverted results relative to the conventional prestack inversion methods.

Time–frequency (T–F) analysis has powerful applications in various fields such as geoscience and engineering. Seismograms are distorted by dispersion and attenuation in such a way that they alter their amplitude and phase spectra. Velocity dispersion is usually neglected in the conventional seismic data processing. However, it has a severe effect on seismic data processing if dispersion is intense in high-attenuation media. Most of the studies on dispersion are theoretical models and laboratory measurements, whereas velocity dispersion analysis directly from field data is rare. In this letter, a method is proposed to estimate the dispersion of vertical seismic profile (VSP) data based on continuous wavelet transform (CWT) in time–frequency domain. The Morlet wavelet, three-parameter wavelet (TPW), and Cauchy wavelet are chosen and compared as mother wavelets in CWT to determine the variations of velocity with frequency. Then, the detailed process of extracting velocity dispersion of propagating waves in a dissipative medium is proposed by using their T–F spectra. Finally, the synthetic and field VSP data are used to demonstrate the validity of the proposed method. The results indicate that the velocity dispersion in synthetic data estimated by TPW and Cauchy wavelet generally match well with the theoretical values while it is not the case for the Morlet wavelet. The velocity dispersion of field data is weak and the estimated velocity basically matches well with the logging data.