Seismic imaging is essential for subsurface exploration, yet achieving high-resolution images remains a significant challenge. This paper investigates the theoretical connections between three resolution-enhancement methods—least-squares migration (LSM), Capon beamforming, and phase correction—within a unified framework based on the Gauss-Markov theorem. This framework reveals their underlying equivalence in enhancing seismic image resolution. LSM iteratively refines seismic images by minimizing data mismatches, but it can be computationally expensive for large-scale datasets. In contrast, Capon beamforming improves resolution by suppressing misaligned noise, while phase correction improves coherence by aligning phase information, both operating as cost-efficient post-processing methods in common image gathers. The theoretical derivations demonstrate the intrinsic equivalence of these methods despite differences in principles and implementations. Synthetic and field data experiments validate the theoretical derivations, demonstrating their interconnections and paving the way for advanced seismic imaging approaches that balance image resolution and computational efficiency.

Stacking plays an essential role in improving the quality of seismic imaging. Conventional stacking is performed by simply averaging the traces within a common-midpoint (CMP) gather after normal-moveout (NMO) correction or a common-image-point gather (CIG) after prestack migration, leading to a stacked trace with limited resolution when seismic signals are not perfectly aligned. This paper presents a modified Capon beamforming algorithm, called Capon stacking, to improve the vertical (temporal or depth) resolution of seismic stacked images. The Capon stacking approach applies a data-dependent weight function to the prestack gather in the lateral direction to filter out misaligned distortions and irregularities in the signal before stacking. The weight function is computed separately for each frequency component using statistical analysis of the NMO-corrected CMP gather or migrated CIG. The implementation of the Capon beamforming method typically involves calculating the inverse of a relatively large-scale covariance matrix, which can be computationally intensive compared to conventional stacking. We propose a fast implementation of the Capon beamforming that yields the same result as the conventional approach, but without performing matrix inversion. Our numerical tests demonstrate that the proposed Capon stacking method is effective and efficient in improving the vertical resolution of seismic images.

COVID-19 has profoundly impacted people's lives, and people are more concerned about medical and health issues, so it is essential to design an efficient method for classifying medical questions.Fine-tuning paradigms based on pre-trained language models have proven effective in recent years.However, PLMs based on fine-tuning paradigms are poorly robust, and there is a gap between the pre-training phase and the downstream task form, resulting in PLMs that cannot use the rich latent knowledge in downstream tasks.We propose a medical question classification method that combines prompt fine-tuning and contrastive learning and uses the large-scale knowledge graph enhancement model ERNIE 3.0 as a feature extractor to address both problems.Our approach utilizes an additional prompt template to enable PLM to unleash the potential in specific tasks and uses a contrast sample strategy to alleviate the problem of confusable samples that are difficult to distinguish.Experiments on a medical question classification dataset show that the method achieves an accuracy of 93.65 percent, with better metrics than recent work.

The pre-trained model in the Chinese text classification task has made significant progress.However, there is a lot of semantically ambiguous and confusing text in the Chinese text, which has a negative impact on the classification model.A triplet network approach for Chinese confusing text classification is proposed to address this problem.This method improves the traditional triplet network's way of randomly constructing sample combinations, compares the feature similarity between the screened confusing text, straightforward text, and ordinary text, and improves the clustering effect of Chinese text features.At the same time, embedding text and label are jointly learned in the same latent space to learn the similarities and differences between texts of the same category and texts of confused categories.Experiments on multiple Chinese text classification datasets demonstrate the negative impact of confusing text on model accuracy and verify the method's effectiveness in this paper.

Seismic migrations are generally formulated as the adjoint operators of linear forward modeling and often lead to images with degraded resolution, unbalanced illumination, and migration artifacts, especially in surveys with geologic complexity and irregular acquisition geometry. Least-squares migration (LSM) is able to mitigate these problems and produce better resolved images that are suitable for subsequent AVO/AVA inversion. However, no matter what domain LSM is implemented in, the computational cost is still several times or even one order of magnitude more than that of the traditional migration. In this article, we present an efficient image-domain least-square Kirchhoff depth migration (LSKDM), in which the Hessian matrix is approximated by a grid of point-spread-functions (PSFs). Traditional PSF computing algorithm requires a nonnegligible cost caused by a successive operation of modeling and migration and has to satisfy a sampling restriction to avoid interference between nearby PSFs. We present in this article that, by using the ray-based Green’s functions and the linear traveltime approximation, the PSFs can be constructed explicitly at a significantly reduced computational cost and are able to adapt flexible spatial sampling that is fine enough to detect small-scale illumination variation. With the constructed PSFs, we formulate an image-domain LSKDM to iteratively solve for the optimal reflectivities. Numerical tests on synthetic and field data examples demonstrate that the proposed LSKDM is highly efficient and is capable of producing images with enhanced spatial resolution and amplitude fidelity when compared with the Kirchhoff depth migration (KDM) image.

In seismic exploration, picking first arrival traveltimes is an important step toward the estimation of subsurface velocity model, which has the direct impact on the well placement. In this letter, we present a 3-D deep learning method for automatic picking. In specific, we employ a supervised 3-D U-shaped full-convolutional network (3-D U-Net) to classify each sample in the 3-D seismic data into two categories: samples before and after the first break. Subsequently, we delineate a surface to separate these two categories, and such surface is corresponding to the desired first arrival traveltimes. In our training phase, besides the original normalized waveform data, we include the energy semblance feature as a channel of the input data, which helps to involve human understanding. The trained 3-D U-Net is applied to both synthetic and real datasets. The synthetic test shows that even if the labels contain some outliers, the predictions are still stable and reasonable in the sense that 3-D U-Net is capable of correcting some incorrect pickings. As for the field dataset test, the predicted results of 3-D U-Net are satisfactory based on human visual verification, and the picked traveltime surface is more consistent compared with the one predicted by the 2-D U-Net.

High-speed train (HST) generates strong and repeatable vibrations that could be used for subsurface imaging and monitoring. Compared with other ambient noise, HST vibrations are generated by a moving source and have striking characteristics as a deterministic source. However, little attention has been paid to the effects of the characteristics of HST sources on the seismic wave retrieval using seismic interferometry. The aim of this study is to investigate what types of seismic waves are retrievable by applying seismic interferometry to HST-induced vibrations. By analyzing the cross correlation of the HST-induced vibrations between two receivers, we find that cross terms are introduced during the cross correlation. These cross terms are nonnegligible for reflection-wave retrieval but can be negligible for the retrieval of direct waves, scattered waves, and refraction waves. This finding has been validated by field data tests. We further demonstrate that the retrieved surface waves can be used to estimate near-surface velocities and that the retrieved scattered surface waves can be used to locate near-surface heterogeneous bodies.