Multiresolution methods are deeply related to image processing, biological and computer vision, scientific computing. The curvelet transform is a multiscale directional that allows an almost optimal nonadaptive sparse representation of objects with edges. It has generated increasing interest in the community applied mathematics signal processing over years. In this article, we present review on transform, including its history beginning from wavelets, logical relationship other...

Seismic velocity is one of the most important parameters used in seismic exploration. Accurate models are key prerequisites for reverse time migration and other high-resolution imaging techniques. Such information has traditionally been derived by tomography or full-waveform inversion (FWI), which consuming computationally expensive, they rely heavily on human interaction quality control. We have investigated a novel method based supervised deep fully convolutional neural network...

Abstract Recently deep learning (DL), as a new data‐driven technique compared to conventional approaches, has attracted increasing attention in geophysical community, resulting many opportunities and challenges. DL was proven have the potential predict complex system states accurately relieve “curse of dimensionality” large temporal spatial applications. We address basic concepts, state‐of‐the‐art literature, future trends by reviewing approaches various geosciences scenarios. Exploration...

Compared with traditional seismic noise attenuation algorithms that depend on signal models and their corresponding prior assumptions, removing a deep neural network is trained based large training set in which the inputs are raw data sets outputs desired clean data. After completion of training, deep-learning (DL) method achieves adaptive denoising no requirements (1) accurate modelings or (2) optimal parameters tuning. We call this intelligent denoising. have used convolutional (CNN) as...

We evaluated a dictionary learning (DL) method for seismic-data denoising. The data were divided into smaller patches, and of patch-size atoms was learned. DL offers more flexible framework to adaptively construct sparse representation according the seismic themselves. being learned from data, did not rely on guess morphology like standard wavelet or curvelet transforms. could learn denoise whether simultaneously in two distinctive steps. Empirical study field showed promising denoising...

A key step in sparsifying signals is the choice of a sparsity-promoting dictionary. There are two basic approaches to design such dictionary: analytic approach and learning-based approach. Although enjoys advantage high efficiency, it lacks adaptivity various data patterns. On other hand, can adaptively sparsify different sets but has heavier computational complexity involves no prior-constraint pattern information for particular data. We have developed double-sparsity dictionary (DSD)...

Machine learning (ML) systems can automatically mine data sets for hidden features or relationships. Recently, ML methods have become increasingly used within many scientific fields. We evaluated common applications of ML, and then we developed a novel method based on the classic support vector regression (SVR) reconstructing seismic from under-sampled missing traces. First, SVR mines continuous hyperplane training that indicates relationship between input with traces output completed data,...

One of the key objectives in geophysics is to characterize subsurface through process analyzing and interpreting geophysical field data that are typically acquired at surface. Data-driven deep learning methods have enormous potential for accelerating simplifying but also face many challenges, including poor generalizability, weak interpretability, physical inconsistency. We present three strategies imposing domain knowledge constraints on neural networks (DNNs) help address these challenges....

We have developed a new algorithm for the reconstruction of seismic traces randomly missing from uniform grid 3D volume. Several algorithms been such reconstructions, based on properties wavefields and signal processing concepts, as sparse representation in transform domain. investigated novel approach, originally introduced noise removal, which is premise that suitable data matrices or tensors, rank (computed by singular value decomposition) increases with traces. Thus, we apply low-rank...

Sparse transforms play an important role in seismic signal processing steps, such as prestack noise attenuation and data reconstruction. Analytic sparse (so-called implicit dictionaries), the Fourier, Radon, curvelet transforms, are often used to represent data. There situations, however, which complexity of requires adaptive transform methods, whose basis functions determined via learning methods. We studied application data-driven tight frame (DDTF) method suppression interpolation...

In this paper, a diffusion-based curvelet shrinkage is proposed for discontinuity-preserving denoising using combination of new tight frame curvelets with nonlinear diffusion scheme. order to suppress the pseudo-Gibbs and curvelet-like artifacts, conventional results are further processed by projected total variation diffusion, in which only insignificant coefficients or high-frequency part signal changed use constrained projection. Numerical experiments from piecewise-smooth textured images...

Restoration/interpolation of missing traces plays a crucial role in the seismic data processing pipeline. Efficient restoration methods have been proposed based on sparse signal representation transform domain such as Fourier, wavelet, curvelet, and shearlet transforms. Most existing are transforms with fixed basis. We considered an adaptive for complex structures. In particular, we evaluated data-driven tight-frame-based regularization method restoration. The main idea tight frame (TF) is...

We propose a convolutional neural network (CNN) denoising based method for seismic data interpolation. It provides simple and efficient way to break though the lack problem of geophysical training labels that are often required by deep learning methods. The new consists two steps: (1) Train set CNN denoisers from natural image clean-noisy pairs learn denoising; (2) Integrate trained into project onto convex (POCS) framework perform alleviates demanding big with similar features as...

Abstract Full‐waveform inversion (FWI) is a powerful geophysical imaging technique that reproduces high‐resolution subsurface physical parameters by iteratively minimizing the misfit between simulated and observed seismograms. Unfortunately, conventional FWI with least‐squares loss function suffers from various drawbacks, such as local‐minima problem human intervention in fine‐tuning of parameters. It particular problematic when applied noisy data inadequate starting models. Recent work...

When we take photos, often get blurred pictures because of hand shake, motion, insufficient light, unsuited focal length, or other disturbances. Recently, a compressed-sensing (CS) theorem which provides new sampling theory for data acquisition has been applied medical and astronomic imaging. The CS makes it possible to superresolution photos using only one few pixels, rather than million with conventional digital camera. Here, further consider so-called deblurring problem: Can still obtain...

In this letter, we apply a new sampling theory named compressed sensing (CS) for aerospace remote to reduce data acquisition and imaging cost. We can only record directly single or multiple pixels while need not the use of additional compression step improve problems power consumption, storage, transmission, without degrading spatial resolution quality pictures. The CS includes two steps: encoding decoding recovery. A noiselet-transform-based single-pixel random Fourier-sampling-based...

In this paper, a detailed implementation of lithium-ion battery life prognostic system using particle filtering framework is presented. A lumped parameter model used to account for all the dynamic characteristics battery: non-linear open-circuit voltage, current, temperature, cycle number, and time-dependent storage capacity. The internal processes are form basis model. Statistical estimates noise in anticipated operational conditions processed provide remaining useful life. then...

In seismic processing, one goal is to recover missing traces when the data sparsely and incompletely sampled. We present a method which treats this reconstruction problem from novel perspective. By utilizing its connection with general matrix completion (MC) problem, we build an approximately low-rank matrix, can be reconstructed through solving proper nuclear norm minimization problem. Two efficient algorithms, accelerated proximal gradient (APG) fitting (LMaFit) are discussed in paper. The...

We have introduced a new decomposition method for seismic data, termed complex variational mode (VMD), and we also designed filtering technique random noise attenuation in data by applying the VMD on constant-frequency slices frequency-offset (f -x) domain. The motivation behind this paper is to overcome potential low performance of empirical (EMD) energy preservation steeply dipping events when used attenuation, resolution signal decomposition. proposed decompose into an ensemble...

Compressed sensing (CS) is a recently introduced concept that enables the recovery of signals sampled below Nyquist rate. A prerequisite for its application sparsity concerned in certain basis. The has been applied to medical imaging, seismic optical remote sensing, and, recently, radar. This paper describes two possible applications CS synthetic aperture radar (SAR): 1) estimation moving-target velocities case high and low signal-to-clutter ratios, as well more than one scatterer single...

Seismic data denoising and interpolation are essential preprocessing steps in any seismic processing chain. Sparse transforms with a fixed basis often used these two steps. Recently, we have developed an adaptive learning method, the data-driven tight frame (DDTF) for interpolation. With its adaptability to data, DDTF method achieves high-quality recovery. For 2D is much more efficient than traditional dictionary methods. But 3D or 5D results high computational expense. The motivation behind...

We introduce a machine learning based method to estimate the P-wave velocity models directly from prestack seismic traces using modified fully convolutional network. The network is tuned map multi-shot models. train with pairs of synthetic and their corresponding which are simulated acoustic wave equations. Multiple shots used as channels in increase data redundancy. training process expensive, but it only occurs once up front. trained then predict testing traces, shows satisfactory...

We have developed an artificial neural network to estimate P-wave velocity models directly from prestack common-source gathers. Our is composed of a fully connected layer set and modified convolutional set. The parameters in the are tuned through supervised learning map multishot gathers models. To boost generalization ability, trained on massive data which natural images that collected online repository. Multishot seismic traces simulated those with acoustic wave equations crosswell...