Distributed Acoustic Sensing(DAS)has many advantages such as full well coverage,high sampling density,and strong tolerance to harsh environments.In recent years,it has made significant contributions to the development of high-precision exploration technology,and is considered to be a new technology that can replace conventional geophones for seismic data acquisition.However,due to the deployment issues of the DAS system,the optic fiber cable may flap and ring along the borehole casing,resulting in a large amount of coupled noise in the acquired seismic data,which seriously affects the subsequent data analysis and interpretation.Existing conventional coupled noise suppression techniques usually require a lot of parameter tuning according to the properties of the noise,such as the noise wavelet,noise period,and noise attenuation characteristics,etc.,which is a time-consuming work and cannot meet the requirements for efficient seismic data processing in the context of big data acquisition.In this paper,we propose a deep learning-based coupled noise suppression method.With the help of the excellent feature extraction ability of Convolutional Neural Network(CNN),the proposed method digs the potential feature differences between the coupled noise and the effective signal,and then realizes the signal-noise separate by implicit processing in the hidden layer of the network according to the feature differences between signal and noise.In addition,through a detailed analysis of the DAS seismic data,we also use forward modeling and parametric modeling to construct a hybrid training set containing real and synthetic DAS data for model optimization.The denoising results of both synthetic DAS data and field DAS data show that the proposed method can effectively suppress the coupled noise with little damage to the effective signal.After processing,the signal that was seriously interfered by coupled noise becomes clearer and more continuous,providing a high-quality data basis for subsequent tasks such as seismic inversion and imaging.

Seismic wavelet interference limits the vertical resolution of seismic data, making it challenging to accurately characterize subsurface geological structures. Seismic reflectivity estimation is a key process for improving the vertical resolution of seismic data. Deep learning-based seismic reflectivity estimation methods typically rely on labels generated from well data, which can be costly to obtain. Moreover, seismic reflectivity estimation is an ill-posed problem, meaning that while the estimated seismic reflectivity may fit the observed data, it can still differ significantly from the true reflectivity, particularly concerning thin layers. To address these challenges, we propose an unsupervised deep learning deconvolution framework guided by a physical convolution model and super-resolution mathematical theory. The network is trained using a combination of reconstruction loss, position prior loss, and sparse loss. Specifically, reconstruction loss establishes a closed-loop connection between the low-resolution seismic data and the estimated seismic reflectivity using the Robinson convolution model, eliminating the need for labels in training. Position prior loss estimates the seismic reflectivity position based on the super-resolution mathematical theory, which is particularly effective for thin layers, improving the interpretability and accuracy of the seismic reflectivity estimation. Sparse loss, based on the L1 norm, enforces sparsity in the seismic reflectivity estimation, enhancing its stability. Both synthetic and field data examples demonstrate that the proposed method outperforms conventional sparse-spike deconvolution method, providing better thin-layer seismic reflectivity estimates and improved lateral continuity.

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.

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 exploration is one of the most widely used geophysical prospecting methods in oil-gas and mineral resources development. Due to the limitations of acquisition conditions, the seismic records are usually contaminated with a large amount of random noise, resulting in a low Signal-to-Noise Ratio (SNR). It seriously affects the identification accuracy of the effective signals, thereby bringing challenges to subsequent inversion and interpretation procedures. In addition, the random noise usually has complex characteristics, such as non-stationary, non-Gaussian and spectral aliasing. The denoising performance for the conventional methods may degrade when confronted with such complex interferences. To achieve the complex noise attenuation, a novel double-layer multi-scale feature fusion denoising network (DMFF-Net) is proposed in this paper. In general, the proposed network has a multi-scale network structure. It utilizes the multi-branch modules to extract the potential features existing in different scales and branches so as to improve the learning ability of the network for complex features of the analyzed seismic data. Meanwhile, we also employ skip connections to fuse the shallow and deep features; then, improve the recover ability of the weak signals. The synthetic and field data processing results indicate that DMFF-Net can suppress the random noise effectively and restore the desired signals accurately. Moreover, it also can significantly improve the SNR. Compared with conventional denoising methods, DMFF-Net has advantages in signal amplitude retention and weak signal recovery.

由于沙漠地区采集环境恶劣、地表地质条件复杂,勘探资料信噪比普遍较低;同时,沙漠区随机噪声与有效信号存在频谱混叠现象,噪声压制难度较大,给后续反演、成像和解释等工作带来了不利影响.近年来,以去噪卷积神经网络(Feed-forward Denoising Convolutional Neural Networks,DnCNN)为代表的深度学习去噪方法已应用于复杂随机噪声抑制,但传统降噪网络一般是根据单一尺度信息提取数据特征,导致针对复杂勘探记录的处理能力可能会下降.为实现沙漠地区复杂噪声的有效衰减,提出一种新型多分支去噪卷积神经网络(Di-verse Branch Block Convolutional Neural Networks,DBBCNN).与传统的 DnCNN 相比,DBBCNN 将不同尺度、不同复杂度的分支结合在一起,丰富了特征空间,并且采用长路径操作融合全局特征和局部特征,提升了网络针对弱信号的特征表达能力.模拟和实际数据实验结果表明,DBBCNN可有效压制沙漠地震资料中的复杂随机噪声,且处理后的记录信噪比显著提升.

Tarim basin mainly composed of desert regions is an important oil and gas exploration area. The desert seismic data acquired from Tarim basin is often characterized by low Signal-to-Noise Ratio (SNR) ; also, the effective signals and noise seriously overlaps in low-frequency domain. These two points bring numerous difficulties to the denoising of desert seismic data, so as to affect the following inversion, imaging, and interpretation. In order to suppress the background noise effectively and recover the effective signals completely, we adopt the basic strategy of Generative Adversarial Network (GAN) and then utilize a denoiser to replace the generator of GAN, so as to propose a novel denoising network for the desert seismic data, named Desert Seismic Convolutional Adversarial Denoising Network (DSCA-Net). In DSCA-Net, we propose a novel loss function by combining the mean square error loss and adversarial loss. Then, this loss function is used to optimize the network parameters of DSCA-Net, so as to obtain the denoising model aiming at the desert seismic data. Synthetic and real experiments show that (1) the proposed DSCA-Net can effectively suppress the desert background noise and significantly enhance the continuity of events ; (2) after processed by DSCA-Net, the signal-to-noise ratio of desert seismic data is obviously improved.

The Central Asian Orogenic Belt (CAOB) is the largest accretionary orogenic belt in the world. The Songnen and Jiamusi blocks, which are located along the eastern segment of the CAOB, form a complex structure influenced by the Mudanjiang ocean closure and the Paleo-Pacific plate subduction. Constraints on the deep structure of the Songnen and Jiamusi blocks are critical for studying the evolution, transformation and current activity of the CAOB. This study presents an analysis and interpretation of a 450-km-long magnetotelluric profile that imaged the deep lithospheric structure of these two blocks. The best-fit three-dimensional resistivity model, based on the nonlinear conjugate gradient algorithm, highlights the key lithospheric structures. The results show that the lithosphere beneath the Lesser Xing'an Range in the Songnen block is characterized by a high-resistivity structure, which is mainly related to the granite formed in multiple stages. The crust forming the Sanjiang basin in the Jiamusi block possesses a high–low–high resistivity structure in the transverse direction, corresponding well to the tectonic units. The low-resistivity structure of the Fujin uplift extends to the mantle, which represents mantle material that intruded the crust. There is a low-resistivity structure beneath the suture zone of the two blocks that is homologous with the large-scale low-resistivity anomaly beneath the basin and is related to the subducted Paleo-Pacific plate. Strong dehydration at the inflection point of the plate reduces the solidus temperature of the mantle material, resulting in melt and upwelling of mantle material. The Jiamusi block can be divided into southern and northern parts, possessing different structures. The southern part represents an original part of either the Gondwana continent or Tarim craton. The Sanjiang basin, in the northern part, has been heavily influenced by the subducted Paleo-Pacific plate, and is a region of strong activity.

Contamination of seismic data by background noise causes difficulties for imaging, reservoir fluid prediction, and stratigraphic interpretation. Desert seismic data poses a particular problem mainly due to two reasons: (1) low signal-to-noise ratio (SNR); (2) serious frequency spectrum overlapping between the effective signals and low-frequency noise (mainly including random noise and surface waves). Therefore, when apply sparse-transform-based methods to denoise desert seismic data, conventional threshold functions fail to distinguish the effective signal coefficients and low-frequency noise coefficients, which is likely to result in residual noise and signal leakage. To solve this problem, we utilize the convolutional neural network (CNN) to act as a threshold function, thereby establishing an optimal non-linear relationship between noisy coefficients and effective signal coefficients. In addition, in order to achieve multi-scale and multi-direction accurate noise suppression, we construct a corresponding training dataset for each sub-band, so as to obtain a CNN-based coefficient selection model suitable for this sub-band. In this paper, we take shearlet transform as an example to verify the effectiveness of the proposed CNN-based threshold function. Synthetic and real examples demonstrate that our method can effectively suppress the desert low-frequency noise and completely recover the effective signals reflected by layers.

The modelling technique contributes to understanding noise nature and properties. Prior work has established a primary random noise model in the homogeneous medium, however, this strict assumption of the medium may be not valid under the actual environment so that will reduce the modelling accuracy. Therefore, in this paper, a random noise model is established in the mixed heterogeneous medium to improve the modelling accuracy, so the scattered mechanism is used to describe the random noise field in the heterogeneous medium. Since the perturbation method is always applied to solve the scattering problem, the noise wave field can be regarded as the superposition of the perturbation wave field and the unperturbation wave field. Consequently, the random noise model reveals that the desert random noise is mainly caused by the wind, and it concentrates on 1-20 Hz. A detailed comparison is made between the noise model established in the homogeneous medium and the proposed noise model, the results illustrate that the proposed noise model is more similar to the actual noise. Besides, for embodying the model's practical application value, the proposed noise model is adopted as the background noise to determine the VMD (variational mode decomposition) denoising parameters. The satisfactory denoising performance supports the usefulness of the proposed noise model for the random noise attenuation.

In tectonics, the east margin of the Songliao basin lies in the east of the Songnen-Zhangguangcai mountain microplate, which has experienced the influences of the regional tectonic stress field, such as the northward extrusion generated by closing of the suture zone in Xilamulun River, and the extruding at the east and north-east side of the domain generated by scissor-type closing from west to Wast of the Mongolia-Okhotsk ocean, as well as the extruding generated by westward subduction of the western Pacific plate, thus resulting in complicated crustal variation. A series of new understandings have been obtained by deep seismic reflection in the Songliao-Drep research. While whether these knowledge are also correct for the eastern margin of the basin remains unclear. In addition, what kind of changes took place in the macro and micro characteristics of the Mohodiscontinuity from the Songliao basin to its east margin and what is the formation mechanism are also needed to be further studied. In order to answer these questions, a west-east deep seismic reflection profile was set up, which extends from the west of Harbin to the nearby Shangzhi city with a length about 150 km. The research shows that there is an obvious crustal difference between the east margin domain and Songliao basin, i. e. from the tri-partition of Songliao basin crust changing into the bipartition of the east margin. The depth of the Moho-discontinuity is about 26 km at the east edge of the profile, and the formation mechanism of its morphology can be explained by the equilibrium theory. A group of large two-way thrusting structures are present in the upper crust, and the main body overridden is speculated to be the palaeo-Asian ocean sedimentary layer, that is the C-P system marine facies layer. These understandings provide a theoretical basis for further investigating Neopaleozoic marine facies strata in Northeast China as well as crustal structure of Northeast Asia.

Desert low-frequency noise is a kind of noise in desert seismic exploration records, with significant low-frequency characteristics. Severe frequency aliasing occurs because the noise is in the same frequency band as the seismic signal. In addition, the interference noise has strong energy over whole time period of seismic records, which makes the signal easily submerged in the noise. These characteristics of desert low-frequency noise challenge traditional denoising methods. Aiming at the noise attenuation of low signal-to-noise ratio (SNR) seismic exploration records in desert areas, first half-quadratic optimization approach is proposed to solve the energy minimization problem instead of common optimization methods in the co-sparse analysis model. And then, shrinkage function is introduced into the model by the additive form of half-quadratic optimization, which makes the model remove the restriction of sparsity promoting function. Finally, according to the characteristics of seismic exploration records, the training data are normalized and then trained. Both the synthetic and real data experiments prove that the improved model can better overcome the frequency aliasing and more thoroughly remove the low-frequency noise compared with the traditional co-sparse analysis model.

One of the difficulties in desert seismic data processing is the large spectral overlap between noise and reflected signals. Existing denoising algorithms usually have a negative impact on the resolution and fidelity of seismic data when denoising, which is not conducive to the acquisition of underground structures and lithology related information. Aiming at this problem, we combine traditional method with deep learning, and propose a new feature extraction and denoising strategy based on a convolutional neural network, namely VMDCNN. In addition, we also build a training set using field seismic data and synthetic seismic data to optimize network parameters. The processing results of synthetic seismic records and field seismic records show that the proposed method can effectively suppress the noise that shares the same frequency band with the reflected signals, and the reflected signals have almost no energy loss. The processing results meet the requirements of high signal-to-noise ratio, high resolution and high fidelity for seismic data processing.

Low-amplitude signal detection is a key procedure in borehole microseismic and desert seismic exploration. Usually, signals are difficult to detect due to their low amplitude and noise contamination. To solve this problem, we propose a method combining shearlet energy entropy with a support vector machine (SVM) to detect low-amplitude signals. In the proposed method, the signal feature is extracted using shearlet energy entropy. The signal is more sparsely represented in the shearlet domain because of the multi-scale and multi-direction characteristic of the shearlet transform, which favours signal feature extraction. Furthermore, in calculating shearlet energy entropy, we use the correlation of shearlet coefficients to enhance the difference between signal and noise in the shearlet domain. Shearlet energy entropy makes the SVM achieve a more accurate classification result compared with other traditional features such as amplitude and energy. The results of synthetic and field data show that our method is more effective than the STA/LTA and the convolutional neural network for low-amplitude microseismic signal and desert seismic signal detection.

Due to the effect of various environment factors, the random noise in desert seismic exploration has complex characteristics, including low frequency, non-Gaussian and frequency band aliasing of signal and noise. Therefore, it is difficult for the denoising processing. Aiming at this problem, a Multi-level Wavelet Convolution Neural Network (MWCNN) is proposed to suppress the desert noise. MWCNN is a combination of two-dimensional discrete wavelet transformation and convolution neural network. Specifically, Discrete Wavelet Transformation (DWT) and inverse wavelet transformation (IWT) are used to replace the pooling layer and up-convolution of U-net respectively. So that the trade-off between receptive field and computational efficiency can be achieved. Consequently, the expansion of the receptive field can obtain more overall information of the events. In this paper, by adjusting the training set and structure of MWCNN. it is applied to suppress the random noise in desert seismic exploration. Furthermore. compared with other neural networks. MWCNN achieves better better denoising effect and better events' continuity by enlarging the receptive field in desert seismic records. And experiments on simulated synthetic records and actual seismic records respectively show our trained MWCNN model achieve a satisfactory denoising performance for the random noise in desert seismic exploration.

The suppression of random noise is a crucial step before seismic data analysis. Random noise in desert areas has the characteristics of low frequency and non-stationary, and there is serious spectrum aliasing between random noise and effective signals, which makes it difficult to suppress such noise. In recent years, some methods based on signal rank minimization have achieved remarkable results in seismic random noise suppression. Since the implementation of low rank matrix approximation is an iterative process, noise estimation is an indispensable step before each iteration, but also an important step. The noise estimation method previously used is to calculate the residuals of the original noisy patch data and the corresponding iterative denoising version, which is intuitively considered as the filtered noise. This method may be very inaccurate in the case of high noise levels or complex seismic records. In this paper, a noise estimation method based on geometric texture is introduced to estimate the noise level by selecting weak textured patches in all seismic texture patches. At the same time, we reduce the loss of effective signals by truncating the singular values in each iteration. Experiments on both synthetic and field seismic data show that this method has better effect on suppressing random noise in desert areas.

SUMMARY The importance of low-frequency seismic data has been already recognized by geophysicists. However, there are still a number of obstacles that must be overcome for events recovery and noise suppression in low-frequency seismic data. The most difficult one is how to increase the signal-to-noise ratio (SNR) at low frequencies. Desert seismic data are a kind of typical low-frequency seismic data. In desert seismic data, the energy of low-frequency noise (including surface wave and random noise) is strong, which largely reduces the SNR of desert seismic data. Moreover, the low-frequency noise is non-stationary and non-Gaussian. In addition, compared with seismic data in other regions, the spectrum overlaps between effective signals and noise is more serious in desert seismic data. These all bring enormous difficulties to the denoising of desert seismic data and subsequent exploration work including geological structure interpretation and forecast of reservoir fluid. In order to solve this technological issue, feed-forward denoising convolutional neural networks (DnCNNs) are introduced into desert seismic data denoising. The local perception and weight sharing of DnCNNs make it very suitable for signal processing. However, this network is initially used to suppress Gaussian white noise in noisy image. For the sake of making DnCNNs suitable for desert seismic data denoising, comprehensive corrections including network parameter optimization and adaptive noise set construction are made to DnCNNs. On the one hand, through the optimization of denoising parameters, the most suitable network parameters (convolution kernel、patch size and network depth) for desert seismic denoising are selected; on the other hand, based on the judgement of high-order statistic, the low-frequency noise of processed desert seismic data is used to construct the adaptive noise set, so as to achieve the adaptive and automatic noise reduction. Several synthetic and actual data examples with different levels of noise demonstrate the effectiveness and robustness of the adaptive DnCNNs in suppressing low-frequency noise and preserving effective signals.

In seismic exploration, random noise is an obstacle to the extraction of the effective signals, so the investigation aimed at random noise is the basis of signal processing. It is of great significance to analyze the noise properties and establish accurate noise models. Since the complex changes of the actual medium seriously affect propagation characteristics, it is necessary to establish a noise model in a more realistic medium. In this letter, we suppose a weakly heterogeneous medium whose properties vary with the position. And the link between the Lam constants of the medium and noise properties is established. Therefore, a wave equation is deduced in that medium to describe the propagation law of desert seismic exploration random noise. Based on the Greens function, the random noise field is obtained by superimposing all wave fields excited by each pointlike source. Afterward, quantitative comparisons between the actual random noise and the proposed random noise model are given. The results manifest that there are significant similarities in mathematical characteristics between them. Moreover, compared with the noise model in the homogeneous medium, the proposed noise model is more reliable. In order to prove the application value of the random noise model, it is first applied to construct a complete training set for denoising convolutional neural networks, which is valuable for attenuating the desert seismic exploration random noise. This is an effective way to extend noise data. Consequently, this feasible application will strongly promote the application of neural networks in seismic exploration.