The lack of low frequency information and a good initial model can seriously affect the success full waveform inversion (FWI), due to inherent cycle skipping problem. Computational extrapolation is in principle most direct way address this issue. By considering bandwidth extension as regression problem machine learning, we propose an architecture convolutional neural network (CNN) automatically extrapolate missing frequencies without preprocessing post-processing steps. bandlimited...

The lack of the low frequency information and good initial model can seriously affect success full waveform inversion (FWI) due to inherent cycle skipping problem. Reasonable reliable extrapolation is in principle most direct way solve this In paper, we propose a deep-learning-based bandwidth extension method by considering as regression Deep Neural Networks (DNNs) are trained automatically extrapolate frequencies without preprocessing steps. band-limited recordings inputs DNNs and, our...

Abstract Seismic wave arrival time measurements form the basis for numerous downstream applications. State‐of‐the‐art approaches phase picking use deep neural networks to annotate seismograms at each station independently, yet human experts seismic data by examining whole network jointly. Here, we introduce a general‐purpose network‐wide algorithm based on recently developed machine learning paradigm called Neural Operator. Our model, Phase Operator, leverages spatio‐temporal contextual...

Full waveform inversion (FWI) strongly depends on an accurate starting model to succeed. This is particularly true in the elastic regime: The cycle-skipping phenomenon more severe FWI compared acoustic FWI, due short S-wave wavelength. In this paper, we extend our work extrapolated (EFWI) by proposing synthesize low frequencies of multi-component seismic records, and use those "artificial" seed frequency sweep FWI. Our solution involves deep learning: separately train same convolutional...

Using large language models (LLMs) integration platforms without transparency about which LLM is being invoked can lead to potential security risks. Specifically, attackers may exploit this black-box scenario deploy malicious and embed viruses in the code provided users. In context, it increasingly urgent for users clearly identify they are interacting with, order avoid unknowingly becoming victims of models. However, existing studies primarily focus on mixed classification human...

In this article, an intelligent pilling prediction model using back-propagation neural network and optimized with genetic algorithm is introduced. Genetic proposed in consideration of the initial weight threshold artificial network, further improves training speed accuracy for polyester–cotton blended woven fabrics. The results show that maximum numbers steps by are reduced from 164 to 137 compared model. fitness improved 0.914 0.945. simulation increased 0.912 0.987. And root mean square...

Computational low frequency extrapolation is in principle the most direct way to address cycle skipping problem full waveform inversion (FWI). We propose a method of extrapolated (EFWI), where FWI allowed make use data augmented by increasing its band with convolutional neural network (CNN). In CNN (EFWI-CNN), low-wavenumber components model are determined from frequencies, before proceeding sweep bandlimited data. The proposed deep-learning low-frequency shows adequate generalizability for...

This paper investigates the 3D domain generalization (3DDG) ability of large models based on prevalent prompt learning. Recent works demonstrate performances point cloud recognition can be boosted remarkably by parameter-efficient tuning. However, we observe that improvement downstream tasks comes at expense a severe drop in generalization. To resolve this challenge, present comprehensive regulation framework allows learnable prompts to actively interact with well-learned general knowledge...

Abstract Fault zones accommodate relative motion between tectonic blocks and control earthquake nucleation. Nanocrystalline fault rocks are ubiquitous in “principal slip zones” indicating that these materials determining stability. However, the rheology of nanocrystalline remains poorly constrained. Here, we show such an order magnitude weaker than their microcrystalline counterparts when deformed at identical experimental conditions. Weakening is hence intrinsic, it occurs once layers form....

SUMMARY Full-waveform inversion (FWI) relies on low-frequency data to succeed if a good initial model is unavailable. However, field seismic excited by active sources are typically band-limited above 3 Hz. By extrapolated FWI, we can start from computational low frequencies data. extrapolation with deep learning challenging for since neural network trained synthetic usually generalizes poorly real Here use semi-supervised method extrapolate training without labels. Specifically, CycleGAN...

Full waveform inversion (FWI) strongly depends on an accurate starting model to succeed. This is particularly true in the elastic regime: The cycle-skipping phenomenon more severe FWI compared acoustic FWI, due short S-wave wavelength. In this note, we extend our work extrapolated (EFWI) by proposing synthesize low frequencies of multi-component seismic records, and use those "artificial" seed frequency sweep FWI. By leveraging deep learning technologies, separately train two neural networks...

PreviousNext No AccessSEG Technical Program Expanded Abstracts 2015Genetic algorithm full waveform inversion for microseismic locationAuthors: Pan Zhang*Liguo HanHan GaoHongyu SunPan Zhang*Jilin UniversitySearch more papers by this author, Liguo HanJilin Han GaoJilin and Hongyu SunJilin authorhttps://doi.org/10.1190/segam2015-5802979.1 SectionsAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InRedditEmail Abstract We present a...

Passive seismic interferometry is a vastly generalized blind deconvolution question, where different paths through the Earth correspond to channels called Green's functions; sources are completely incoherent and not shared by channels, question estimate (channels) that present in dataset. SI, turning noise signal, has numerous applications, from monitoring industrial activities crustal structure investigation. No standard method of signal processing will solve SI. Instead, domain scientists...

Recently, a growing number of work design unsupervised paradigms for point cloud processing to alleviate the limitation expensive manual annotation and poor transferability supervised methods. Among them, CrossPoint follows contrastive learning framework exploits image data understanding. Although promising performance is presented, unbalanced architecture makes it unnecessarily complex inefficient. For example, branch in ~8.3x heavier than leading higher complexity latency. To address this...

Earthquake hypocenters form the basis for a wide array of seismological analyses. Pick-based earthquake location workflows rely on accuracy phase pickers and may be biased when dealing with complex sequences in heterogeneous media. Time-reversal imaging passive seismic sources cross-correlation condition has potential high resolution, but carries large computational cost. Here we present an alternative deep-learning approach by combining benefits neural operators wave propagation time...

Seismic wave arrival time measurements form the basis for numerous downstream applications. State-of-the-art approaches phase picking use deep neural networks to annotate seismograms at each station independently, yet human experts seismic data by examining whole network jointly. Here, we introduce a general-purpose network-wide algorithm based on recently developed machine learning paradigm called Neural Operator. Our model, PhaseNO, leverages spatio-temporal contextual information pick...

Abstract The pBRDF model is able to relate the properties of target materials polarization information incident and reflected light, an important basis for obtaining targets in space. It detection space targets. P-G first strictly officially released, but there are still deficiencies. In this paper, we analyze assumption framework model, derive imperfections through analysis framework, add scattering phase function enhance existing model. On parameter inversion, output results compared with...

Under the assumptions of diffuse wavefields or energy equipartitioning, theoretical studies showed that Green's function can be retrieved from cross-correlation ambient noise in seismic interferometry (SI). However, practice, correlograms are not equal to empirical since for correlation-based SI generally satisfied realistic situations. In framework supervised learning, we propose train deep neural networks overcome two limitations SI: temporal limitation passive recordings, and spatial...

We propose a surface-wave analysis method, extrapolated dispersion inversion (EDI), to image the near-surface shear-wave velocity structures beyond penetration depth of conventional methods. Active-source surface waves are main type seismic data for an imaging less than one kilometer. The relatively low-frequency play important role in inversion, by increasing investigation and decreasing uncertainty. However, recorded from active source generally lack components below 3 Hz since acquiring...

We found that the direct arrival has "great relation" to signal-to-blending noise ratio (S/N) during deblending of simulated and real simultaneous source dataset. In this paper, influence was presented, a new method attenuate in debending process proposed at same time. The example synthetic datasets proved S/N deblended shot gather will be doubled. Presentation Date: Tuesday, October 18, 2016 Start Time: 4:10:00 PM Location: Lobby D/C Type: POSTER