Random Noise Attenuation Research Articles

Residual learning with feedback for strong random noise attenuation in seismic data

In random seismic noise attenuation, when the noise energy is higher than or close to a signal, it is difficult to distinguish the signal from the noise. This random noise is relatively strong compared to the signal and is called strong random noise. We have developed a deep learning framework to recover the signal from the strong random noise. The framework is based on a residual learning network and feedback connection and is called the feedback residual network. The residual network (ResNet) suppresses random noise through residual fitting and improves the network’s training efficiency. The feedback connection allows the framework to process data in iterations. In each iteration, the feedback connection proportionally combines the input and output of the ResNet to reconstruct a new input with a lower noise level. This enhances denoising performance by asymptotically decreasing the input noise level and retrieving the remaining signals from the estimated noise, thereby reducing the difficulty of strong random noise attenuation. We terminate the feedback iterations according to the energy change of the estimated noise in each iteration. Synthetic and field examples demonstrate that our network can effectively attenuate the strong random noise.

Random Noise Attenuation Using an Unsupervised Deep Neural Network Method Based on Local Orthogonalization and Ensemble Learning

Random Noise Attenuation Using an Unsupervised Deep Neural Network Method Based on Local Orthogonalization and Ensemble Learning

Reinforcement Learning-Based Denoising Model for Seismic Random Noise Attenuation

The random noise attenuation is an essential step in seismic data processing. Due to complex geological conditions and acquisition environment, the intensity of effective signal and random noise varies in time and space. Additionally, the morphology of seismic event is complex and diverse, such as large dips and fast changes. These complex conditions necessitate that the denoiser adjust filtering policy dynamically. In this paper, we propose a reinforcement learning-based seismic denoising (RLSD) model with the framework of asynchronous advantage actor-critic (A3C). In A3C framework, the RLSD agent utilizes a policy network to learn a denoising policy for the state that is the sample of seismic data and selects a suitable filter from the preset action space composed of multiple simple and effective seismic filters with different parameters. Moreover, the RLSD agent utilizes a value network and a region-adaptive weighted reward function to accurately evaluate the denoising effect of nonstationary seismic signals. A curriculum learning approach is adopted to achieve convergence of the proposed RLSD model under complex seismic data by training from the stationary training data to nonstationary training data, and make the model more suitable to the data to be processed by using a local similarity-based reward function to fine-tune the model. The synthetic and field seismic data applications confirm that the proposed RLSD model achieves a significant performance in preserving nonstationary signals and suppressing noise by adaptively adjusting the denoising policy according to the complex structural features and noise levels. The source code is available on https://github.com/liangc-code/RLSD.

Unsupervised Seismic Random Noise Suppression Based on Local Similarity and Replacement Strategy

Improving the signal-to-noise ratio and suppressing random noise in seismic data is critical for high-precision processing. Although deep learning-based algorithms have gained popularity as denoising methods, they suffer from poor generalization ability, resulting in high training set construction cost and computation cost. To address this problem, we propose an unsupervised learning-based denoising method that includes an improved denoising strategy based on local similarity and replacement, a corresponding training method, and an improved network based on UNet. Our training method takes advantage of network convergence and allows direct training on the test region, effectively solving the problems associated with denoising methods using generalization ability while improving training performance. In addition, our network is specifically designed for the training method and incorporates various improvements that could further enhance the training effectiveness. Our method outperforms traditional denoising methods, as demonstrated by tests on synthetic and field data, with superior performance in random noise attenuation and reflection event reconstruction.

Attenuation of Seismic Random Noise With Unknown Distribution: A Gaussianization Framework

Attenuation of Seismic Random Noise With Unknown Distribution: A Gaussianization Framework

A Self-Supervised Denoising Method Based on Deep Noise Estimation

Seismic random noise attenuation is an essential procedure when processing seismic data. Due to various acquisition environments and complex geological conditions, random noise in seismic field data exhibits spatiotemporal levels, significantly increasing the difficulty of extracting seismic signals. The deep learning (DL) methods have shown excellent performances for seismic denoising. However, most existing discriminative DL methods cannot fit field data with noise levels and signal structures that differ from the training set. To tackle the unmatching challenge, we propose a self-supervised deep denoising model named noise estimation-based convolution neural network (NE-CNN), which contains a multiscale denoising module as a pretrained model and a dual-path noise estimation module that estimates the noise level of each data patch with a generalized Gaussian distribution and gray-level co-occurrence matrix (GLCM). With Stein’s unbiased risk estimate (SURE), we can fine-tune the pretrained denoising module according to the estimated noise levels in a self-supervised style solely on data to be processed, leading to a boost of robustness to nonstationary seismic noise. In synthetic and field data tests, NE-CNN performed satisfactorily compared with discriminative DL methods in denoising effects on seismic data with nonstationary noise.

Random Noise Attenuation of Seismic Data via Self-Supervised Bayesian Deep Learning

Random Noise Attenuation of Seismic Data via Self-Supervised Bayesian Deep Learning

S2S-WTV: Seismic Data Noise Attenuation Using Weighted Total Variation Regularized Self-Supervised Learning

Seismic data often undergoes severe noise due to environmental factors, which seriously affects subsequent applications. Traditional hand-crafted denoisers such as filters and regularizations utilize interpretable domain knowledge to design generalizable denoising techniques, while their representation capacities may be inferior to deep learning denoisers, which can learn complex and representative denoising mappings from abundant training pairs. However, due to the scarcity of high-quality training pairs, deep learning denoisers may sustain some generalization issues over various scenarios. In this work, we propose a self-supervised method that combines the capacities of deep denoiser and the generalization abilities of hand-crafted regularization for seismic data random noise attenuation. Specifically, we leverage the Self2Self (S2S) learning framework with a trace-wise masking strategy for seismic data denoising by solely using the observed noisy data. Parallelly, we suggest the weighted total variation (WTV) to further capture the horizontal local smooth structure of seismic data. Our method, dubbed as S2S-WTV, enjoys both high representation abilities brought from the self-supervised deep network and good generalization abilities of the hand-crafted WTV regularizer and the self-supervised nature. Therefore, our method can more effectively and stably remove the random noise and preserve the details and edges of the clean signal. To tackle the S2S-WTV optimization model, we introduce an alternating direction multiplier method (ADMM)-based algorithm. Extensive experiments on synthetic and field noisy seismic data demonstrate the effectiveness of our method as compared with state-of-the-art traditional and deep learning-based seismic data denoising methods.

Structure-Oriented CUR Low-Rank Approximation for Random Noise Attenuation of Seismic Data

Structure-Oriented CUR Low-Rank Approximation for Random Noise Attenuation of Seismic Data

Total Variation Regularized Self-Supervised Bayesian Deep Learning for Seismic Random Noise Attenuation

Total Variation Regularized Self-Supervised Bayesian Deep Learning for Seismic Random Noise Attenuation

Self-Supervised Seismic Random Noise Attenuation With Spatial Attention From a Single Section

Self-Supervised Seismic Random Noise Attenuation With Spatial Attention From a Single Section

Random Noise Attenuation by Self-supervised Learning from Single Seismic Data

Random Noise Attenuation by Self-supervised Learning from Single Seismic Data

Random noise attenuation using the novel Estimated Noise Pattern Denoising Algorithm

This paper introduces the estimated pattern denoising (EPD) wavelet transform for random noise attenuation in geophysical data. The proposed approach combines the capability of the Gaussian filter and dual-tree rational dilation wavelet transform (DT-RADWT) in random noise detection and suppression; we called this method Estimated Pattern Denoising (EPD). The EPD is an innovative approach in terms of estimation of the location and amplitude of the noise pattern, directly from the data. The employed approach produces a higher quality factor (Q-factor) than the conventional dyadic discrete wavelet transform (DWT) and separates the noise from the signal with higher accuracy. The EPD provides a data-driven scheme that resolves the complexity of the random noise model in noise suppression, using an auxiliary Gaussian filter. This approach does not require prior information about the noise source, statistical distribution, or frequency range. We show successful suppression of random noise using the proposed approach on synthetic and real field data.

Seismic data interpolation using nonlocal self-similarity prior

The use of a nonlocal self-similarity (NSS) prior, which refers to each reference patch always having many nonlocal similar patches, has demonstrated its effectiveness in seismic data random noise attenuation because of the repetitiveness of textures and structures in their global position. However, NSS-based approaches face challenges when dealing with seismic interpolation. In the presence of missing traces, similar patch matching may be highly unreliable, resulting in a limited performance of interpolation. To solve this problem, a two-stage iterative seismic-interpolation framework based on a rank-reduction (RR) algorithm is developed. In the first stage, preinterpolation seismic data are used to guide the similar patch matching, and the problem of missing trace recovery for the stacked matched patches is converted to the problem of low-rank matrix completion. In the second stage, the similar patches are directly searched on the interpolation result after stage 1 without external help; that is, exploiting its own NSS, which can achieve enhanced interpolation performance. For each iteration, we obtain accurate similarly matched groups and apply a simple and efficient truncated singular value decomposition for RR. Owing to the unique construction method of a low-rank matrix formed by similar patches, our approach can handle irregularly or regularly sampled seismic data. Numerical experiments verify the effectiveness of our method, compared with the curvelet, low-rank matrix fitting, and f- x prediction filtering methods.

Random and coherent noise attenuation for 2D land seismic reflection line acquired in Iraq

ABSTRACT Noises are common events in seismic reflection data that have very striking features in seismograms, affecting seismic data processing and interpretation. Noise attenuation is an essential phase in seismic processing data, usually resulting in enhancing the signal-to-noise ratio (SNR) and improving the subsurface seismic image. Groundroll presence is a major fashion of significant noise in land seismic surveys. It is a type of coherent noise present in seismograms that appears as linear events, in most cases overlapping the reflections and probably making it challenging to recognise. There are several domains used in noise attenuation techniques. The domain transformations are a complex algorithm used commonly during the processing of seismic data; Therefore, a large number of methods have been developed to attenuate these types of noise to preserve the frequency bandwidth and enhance the SNR of the seismic data. In the time-offset domain, the noise wave such as groundroll and random noises overlap over time; a different domain makes it easier to successfully isolate coherent, random noise and reflection events. We have used effective algorithms in different domains such as (shot, receiver, t-x, f-x, f-k, R-T, and offset class) to attenuate coherent and random noises present in the data. The results indicate that the different domains can reveal features and geological structures that have been masked by the noises present in current data. Because these filtering techniques encourage significant improvements in the final image quality in the 2D seismic section, possibly giving the interpreter an advantage, particularly in structural and stratigraphic interpretation.

Random noise attenuation with weak feature preservation via total variation regularization

Dictionary learning (DL) methods have been successfully applied to seismic data denoising. However, when the noise is strong, weak signals cannot be well preserved. Considering the characteristics of seismic data in the spatial-time domain, we propose a dictionary learning method with total variation regularization (DLTV). The DLTV method consists of two steps. The first step is to learn the dictionary; it uses the sparse feature of the data in the dictionary domain and the learned dictionary contains data features. The second step is to use the augmented Lagrange multiplier method to restore the data, and total variation regularization to preserve the weak signal by sampling data information around it. An example with seismic data shows that the proposed method retains the weak features better than the traditional F-X deconvolution (FX-Decon) method and the data-driven tight frame method (DDTF).

Random Noise Attenuation in Tunnel Based on EMD-T-FSS

Random Noise Attenuation in Tunnel Based on EMD-T-FSS

Combined multi-branch selective kernel hybrid-pooling skip connection residual network for seismic random noise attenuation

Abstract To improve the generalization ability of the single pooling (average or maximum pooling) skip connection residual network (SSN) for seismic random noise attenuation, we present a hybrid-pooling skip connection residual network (HSN). In HSN, the hybrid pooling consists of average and maximum pooling and aims to simultaneously capture the local and global features well, ultimately improving the detail recovery capability of HSN. To further improve the network performance and denoising ability of HSN, we propose a combined multi-branch selective kernel (CSK) hybrid-pooling skip connection residual network, which is referred to as CHSN. In CHSN, CSK consists of a three-branch selective kernel (TSK) and our suggested four-branch selective kernel (FSK), and aims to adaptively capture feature maps for high-accuracy effective information recovery. The superior random noise attenuation ability of CHSN is demonstrated in both synthetic three- and actual two-dimensional seismic data.

The 3D conical Radon transform for seismic signal processing

Recently, the attenuation of highly aliased broadband surface waves has received significant attention as their high-frequency and low-speed components render f- k x- k y filters ineffective. As a popular tool widely used in seismic signal processing, the 3D linear Radon ([Formula: see text]) transform does not match the surface wave events in the time domain. This leads to the dispersal of energy over ellipses on the zero-intercept slice, whereas the reflected events over the ellipses are reflected on each intercept slice of the 3D linear Radon transform (LRT) domain, decreasing the resolution in slowness. We have introduced a new type of Radon transform defined on circular cones called 3D conical Radon transform (CRT) for seismic signal processing. Unlike LRT, the CRT maps seismic data to surface integrals on circular cones in the 3D seismic records. Consequently, surface wave events are focused as points on the zero-intercept slice, whereas the reflected events are points on each intercept slice of the CRT domain, which significantly improves the resolution in slowness compared with the LRT. We determine the performance of the CRT for seismic signal processing for random noise attenuation, primary and multiple separations, and surface wave attenuation. Based on a comparison with the LRT, synthetic data and field data sets validate the effectiveness of the CRT method at improving the focusing properties.

Denoising Seismic Data via a Threshold Shrink Method in the Non-Subsampled Contourlet Transform Domain

In seismic exploration, effective seismic signals can be seriously distorted by and interfered with noise, and the performance of traditional seismic denoising approaches can hardly meet the requirements of high-precision seismic exploration. To remarkably enhance signal-to-noise ratios (SNR) and adapt to high-precision seismic exploration, this work exploits the non-subsampled contourlet transform (NSCT) and threshold shrink method to design a new approach for suppressing seismic random noise. NSCT is an excellent multiscale, multidirectional, and shift-invariant image decomposition scheme, which can not only calculate exact contourlet transform coefficients through multiresolution analysis but also give an almost optimized approximation. It has better high-frequency response and stronger ability to describe curves and surfaces. Specifically, we propose to utilize the superior performance NSCT to decomposing the noisy seismic data into various frequency sub-bands and orientation response sub-bands, obtaining fine enough transform high frequencies to effectively achieve the separation of signals and noises. Besides, we use the adaptive Bayesian threshold shrink method instead of traditional handcraft threshold scheme for denoising the high-frequency sub-bands of NSCT coefficients, which pays more attention to the internal characteristics of the signals/data itself and improve the robustness of method, which can work better for preserving richer structure details of effective signals. The proposed method can achieve seismic random noise attenuation while retaining effective signals to the maximum degree. Experimental results reveal that the proposed method is superior to wavelet-based and curvelet-based threshold denoising methods, which increases synthetic seismic data with lower SNR from −8.2293 dB to 8.6838 dB, and 11.8084 dB and 9.1072 dB higher than two classic sparse transform based methods, respectively. Furthermore, we also apply the proposed method to process field data, which achieves satisfactory results.