An accurate seismic facies analysis (SFA) can provide insight into the subsurface sedimentary facies and has guiding significance for geological exploration. Many machine learning algorithms, including unsupervised, supervised, and deep learning algorithms, have been developed successfully for SFA over the past decades. However, SFA and facies classification are still challenging tasks due to the complex characteristics of geological and seismic data. A multiattribute SOM-K-means clustering algorithm, which implements a two-stage clustering by using multiple geological attributes, is proposed and applied for SFA. The proposed algorithm can effectively extract complementary features from the multiple attribute volumes and comprehensively use the different attributes to improve the recognition ability of seismic facies. Experimental results show that the proposed algorithm improves clustering accuracy and can be used as an effective and powerful tool for SFA.

Road traffic noise is a special kind of high amplitude noise in seismic or acoustic data acquisition around a road network. It is a mixture of several surface waves with different dispersion and harmonic waves. Road traffic noise is mainly generated by passing vehicles on a road. The geophones near the road will record the noise while receiving the seismic signal. The amplitude of the traffic noise is much larger than the signal, which masks the effective information and degrades the quality of acquired data. At the same time, the traffic noise is coupled with the effective signal, which makes it difficult to separate them. Therefore, attenuating traffic noise is the key to improving the quality of the final processing results. In recent years, denoising methods based on convolution neural networks (CNN) have shown good performance in noise attenuation. These denoising methods can learn the potential characteristics of acquired data, thus establishing the mapping relationship between the original data and the effective signal or noise. Here, we introduce a method combining UNet networks with asymmetric convolution blocks (ACBs) for traffic noise attenuation, and the network is called the ACB-UNet. The ACB-UNet is a supervised deep learning method, which can obtain the distribution characteristics of noise and effective signal through learning the training data and then effectively separate the two to achieve noise removal. To validate the performance of the proposed method, we apply it to synthetic and real data. The data tests show that the ACB-UNet can obtain good results for high amplitude noise attenuation and is practical and efficient.

Accurately upgoing and downgoing wavefield separation is a critical step in vertical seismic profile (VSP) data processing, as its accuracy is directly related to the imaging quality of VSP data. Traditional methods are based mainly on transforms, and their windows are manually set in the transformed domain to obtain the target wavefield. The manual operations often cause errors and affect the accuracy of wavefield separation. In contrast, deep learning algorithms are more automatic, and have achieved a lot in seismic data processing. We propose to employ a conditional generative adversarial network (CGAN) for wavefield separation in VSP data. A CGAN consists of two main components: a generator, which generates new data samples, and a discriminator, which evaluates the generated samples against real data, with both components trained simultaneously in an adversarial manner to improve the quality of generated samples. The full wavefield serves as a constraint to link the generator and discriminator, ensuring that the separated up- and downgoing wavefields align better with the full wavefield. An asymmetric convolution block is introduced to more effectively capture the directional features of the VSP wavefield. To mitigate the influence of amplitude differences between the waves on the network update, the relative downgoing wavefield (obtained by subtracting the predicted upgoing wavefield from the full wavefield) is included in the loss function. Numerical experiments demonstrate that the trained network can effectively learn the characteristics of the up- and downgoing waves, especially their propagation directions, achieving high-precision wavefield separation.

The limitations of the coverage range and density of transmission wave often result in less-than-ideal results in traveltime tomography. By contrast, joint transmission-reflection traveltime tomography can not only recover deep structures that transmission tomography cannot detect, but also optimize its inversion results. In this article, we perform joint tomography on borehole seismic (vertical seismic profile, reverse vertical seismic profile, and crosswell seismic) data to obtain near-wellbore structures. In the first part, we solve the factored eikonal equation by the discontinuous Galerkin (DG) method to calculate the transmission/reflection traveltime. Owing to the large wavefront curvature near the source point, the traveltime errors generated by the numerical simulation will propagate from the source to all the calculation domains. According to the factorization principle, the equation solution is decomposed into two parts to solve the point-source singularity. To further improve the accuracy of solving traveltime, we use the DG method to solve the factored eikonal equation with additive factors (the factored DG method), obtaining second-order accuracy solution. The adjoint-state method is employed in the inversion section to calculate the gradient of the misfit function. We use the traveltime difference observed inside the model to define the misfit function, which is more suitable for borehole seismic and avoids the influence of surface normal vectors on gradients. Numerical tests applied on models indicate that the joint tomography method has the potential to accurately inverse the seismic structure information near the well and recover the deep underground structure.