Velocity model inversion is one of the most challenging tasks in seismic exploration, and an accurate velocity model is essential for high-resolution seismic imaging. Recently, velocity inversion methods based on deep learning (DL), particularly convolutional neural networks (CNNs), have attracted considerable attention from the seismic exploration community. These researchers aim to directly estimate the velocity model from raw seismograms using a well-trained model. Although CNN-based velocity inversion methods have demonstrated remarkable performance in terms of intelligence and automation, their inversion performance is often constrained by a limited long-range dependence. Specifically, when conducting a convolutional operation on raw seismic data using small kernels (i.e., 1 × 1, 3 × 3, 5 × 5, and 7 × 7), CNN-based methods extract only the local features and neglect the weak spatial correlation between different local features that reflect the information of the same interface. This correlation could assist CNN in providing an overview of the seismic data and promote inversion performance when using DL. Furthermore, the time-varying properties of seismic data pose a challenge to the weight sharing of CNNs. Here, we have developed a new DL framework based on a transformer, called the seismic velocity inversion transformer (SVIT), to address the problem of velocity inversion. SVIT uses a self-attention mechanism to capture the long-range dependence of seismic data, rather than stacking multiple convolutional layers as in CNNs. Thus, SVIT can provide more informative remote features for building velocity models. The validity and reliability of the proposed method are demonstrated through numerical experiments using synthetic models. Compared with the conventional full-waveform inversion method and an existing CNN-based velocity inversion method, our SVIT indicates greater consistency with the target in terms of the velocity value, subsurface structures, and geologic interfaces and is expected to provide a new DL-based solution to resolve inversion problems.

Seismic background noise often damages the desired signals, thereby resulting in some artifacts in the seismic imaging that follows. Since about 2016, some supervised-deep-learning methods have shown impressive performance in seismic data denoising, but they usually only consider single-scale features and neglect the multi-scale strategy. To further reinforce their denoising performance, a novel multi-scale convolutional neural network (CNN) combined with spatial attention mechanism, called multi-scale spatial attention denoising network (MSSA-Net), is proposed to tell weak reflected signals apart from strong seismic background noise. Unlike conventional single-scale CNNs, this proposed MSSA-Net can achieve the extraction of multi-scale features which is beneficial for the suppression of strong noise and the recovery of weak reflected signals. Specifically, MSSA-Net contains a principal denoising network and two auxiliary networks. The former utilizes the widen convolution composed of multiple parallel convolution layers with different kernel sizes to capture the informative multi-scale features; the latter two leverage up and down sampling to extract local fine and global coarse features, respectively. Furthermore, a spatial attention block is adopted to fuse these multi-scale features, thereby distinguishing weak reflected signals from strong seismic background noise. Multiple experiments of synthetic and real seismic records demonstrate the effectiveness of MSSA-Net. In addition, compared with two classical single-scale CNNs, MSSA-Net performs better in signal recovery, indicating the positive effect of multi-scale strategy.

Recently, distributed optical fiber acoustic sensing (DAS) is regarded as a transformative technology in seismic exploration. However, both various complex background noise and weak desired signals significantly limit its practical application. To explore an effective denoising method for the vertical seismic profile (VSP) record received by DAS, we propose an improved residual encoder–decoder deep neural network (RED-Net) enhanced by deep iterative memory block (DMB) and channel aggregation block (CAB), called residual channel aggregation encoder–decoder network (RCEN). Here, DMB uses the weight accumulation theory to improve the feature extraction ability and achieve accurate noise elimination. Meanwhile, CAB, using the multi-channel analysis architecture, enhances the weak signal retention performance. In addition, we leverage both the synthetic data obtained by forward modeling and real DAS noise data to construct a sufficient training dataset with high authenticity, thereby meeting the requirement of network training. Both the synthetic and field DAS-VSP data processing results demonstrate the advantage of RCEN compared with competing algorithms, including singular value decomposition (SVD), conventional RED-Net, and feed-forward denoising convolutional neural network (DnCNN).

Distributed optical fiber acoustic sensing (DAS) is an emerging acquisition technology in seismic exploration. However, DAS records are always affected by the complex background noise, resulting in a low signal-to-noise ratio (SNR). In addition, the DAS background noise has different properties from the noise existing in conventional seismic data. Thus, conventional denoising methods may degrade the record when dealing with complex DAS data. To improve the denoising capability, a novel denoising network, called residual modular cascaded heterogeneous network (RMCHN) is proposed. In general, the network is based on the idea of heterogeneous convolution and modular convolutional neural networks. Specifically, different modules are designed to extract the discriminatory features of the DAS data through effective information integration. On this basis, heterogeneous convolution combined with long and short path feature learning strategy is employed to fuse the captured features, thereby improving the feature expression capability and avoiding the information loss. Both synthetic and field denoising results indicate that RMCHN can suppress the DAS background noise with excellent performance in signal restoration, even for the weak signals form deep strata.

Seismic imaging can be solved as an inversion problem, which can be implemented as a least-squares migration (LSM). Compared with conventional migration algorithms, an LSM can produce imaging results with enhanced illumination and resolution. However, solving an inversion problem faces difficulties in convergence, stability, and computational efficiency. To address these issues, efforts have been spent on examining the key elements in an LSM, including the modeling operators, the migration operators, the inversion solvers, etc. Advanced modeling operators are developed to accurately simulate the seismic data. Innovative migration algorithms are implemented to precisely compute the subsurface image. Fast inversion solvers are used to improve computation efficiency. Although all these techniques are robust to improve the LSMs, they often are studied independently. It is not often considered whether these elements are properly combined when used in an LSM. We have constructed LSMs with adjoint modeling and migration operators, and we develop algorithms to prepare input shot data for these LSMs.

Seismic data reconstruction is always an essential step in the field of seismic data processing. Effective reconstruction methods can obtain high-density information at low-cost and also recover missing seismic data. Due to the strong feature extraction ability, convolutional neural network (CNN) has shown remarkable performance in numerous fields of data processing and been gradually applied to seismic data reconstruction. However, most of CNN-based methods applied to seismic data reconstruction only consider features in single scale or just utilize simple interactions between different scales, which is likely to result in performance degradation when facing complex and extremely incomplete seismic data. To further promote the performance of CNN-based methods in seismic data reconstruction, a novel multiscale enhanced attention network (MSEA-Net) is proposed based on the self-enhanced scheme. In general, MSEA-Net has a multiscale architecture which can significantly improve the processing accuracy by fusing the potential features in different-resolution seismic data. From the basis, a parallel sparse residual block is designed and applied in MSEA-Net to enhance processing efficiency and avoid overfitting issues. In addition, a dense spatial attention block is also introduced to the network to further reinforce the effective features, thereby strengthening the reconstruction performance. Experimental results demonstrate that our proposed network can effectively reconstruct incomplete seismic data including regular missing data, irregular missing data, and even consecutively missing data with big gap, which is superior than exist interpolation methods including commonly used U-Net.