SUMMARY Porosity characterization is of profound significance for seismic inversion and hydrocarbon prediction. Although semi-supervised learning (SSL) based methods have been used to boost prediction accuracy lateral continuity supervised (SL) inverted subsurface properties, their variations are relatively limited since the relationships between data parameter model straightforward in most reported cases. To further figure out essential differences, we proposed SSL-based network (SSLBN)...

Low-frequency information is important in reducing the nonuniqueness of absolute impedance inversion and for quantitative seismic interpretation. In traditional model-driven methods, low-frequency background from an initial model almost unchanged during process. Moreover, results are limited by quality modeled data extracted wavelet. To alleviate these issues, we have investigated a double-scale supervised method based on gated recurrent encoder-decoder network (GREDN). We first train...

Noise attenuation has been a long-standing but still active topic in seismic data processing. The deep convolutional neural networks (CNNs) have recently adopted to remove the learned random noise from noisy data, it is difficult improve generalization ability of denoisers due limited diversity training sets. In this letter, we investigate an end-to-end denoising CNNs (DCNNs) with novel generation method involving multidimensional geological structure features for denoising. To learn...

Picking first-break (FB) from seismic trace is an important step for refraction and reflection exploration. Conventional picking methods are mostly based on identifying the differences between signal noise in terms of amplitude, phase, or frequency. We investigate a waveform classification FB-picking method fully convolutional neural networks (FCNs) transfer learning (TL). consider as binary image segmentation problem labelling 2D with ones zeros noise. Through FCNs, we achieve fast...

The systematic evaluation of speech separation and enhancement models under moving sound source conditions typically requires extensive data comprising diverse scenarios. However, real-world datasets often contain insufficient to meet the training requirements models. Although synthetic offer a larger volume data, their acoustic simulations lack realism. Consequently, neither nor effectively fulfill practical needs. To address these issues, we introduce SonicSim, toolkit de-designed generate...

ABSTRACT Deep learning has shown excellent performance in simulating complex nonlinear mappings from the seismic data to elastic parameters. However, acoustic impedance estimated a direct mapping waveform P‐wave (single‐input network) is hampered by limited frequency bands. In this paper, we propose incorporate low‐frequency model constrain inversion (multi‐input network). We add feature fusion layer force lateral smoothness. Besides, usually, given survey likely contain only few well logs,...

The picking efficiency of seismic first breaks (FBs) has been greatly accelerated by deep learning (DL) technology. However, the accuracy and DL methods still face huge challenges in low signal-to-noise ratio (SNR) situations. To address this issue, we propose a regression approach to pick FBs based on bidirectional long short-term memory (BiLSTM) neural network implicit Eikonal equation 3D inhomogeneous media with rugged topography target region. We employ regressive model that represents...

Summary Seismic impedance inversion plays an important role in fine characterization of lithology and reservoir prediction. The conventional methods cannot generate low-frequency information during the process. However, components are highly significant reducing multi-solution results for quantitative interpretations. To obtain impedance, interpretable gated recurrent encoder-decoder networks (GRED) dual-driven is proposed. We consider two supervisors including well-log curves corresponding...

First-break (FB) picking can provide traveltime information that is useful for seismic exploration. Due to the limitation of actual acquisition conditions, receiving sensors are often irregularly distributed, which cannot a complete FB study area. The fully from data with partially missing traces and/or poor-quality generally needs interpolation before picking. Moreover, most conventional methods ignore spatial correlation FBs, and require inputting sensitive attributes extracted data....

Summary The classical model-driven seismic high-resolution processing method using (time-variant) deconvolution or reflectivity inversion is derived to be a special case of the data-driven artificial neural networks (ANNs). To obtain subsurface images, we propose an interpretable gated recurrent encoder-decoder (IGREDN), advanced ANN-type method, process observed band-limited data. developed consist encoding network and decoding network, which are both mainly composed bidirectional unit...

Summary Pre-stack three-parameter inversion commonly refers to estimating the three elastic parameters of longitudinal velocity VP, transverse VS and density ρ from observed seismic angle gather. Conventional deep learning-based methods have significant lateral discontinuities neglect correlation between elasticity parameters. In this abstract, we investigate a neural network under low-frequency information constraint for simultaneous inversion, which realizes related tasks based on idea...

Summary The estimation of impedance and gas saturation from the same subsurface reservoir is historically implemented by two consecutive workflows: seismic inversion to derive petrophysical modeling further transform into saturation. To realize simultaneous prediction for saturation, we propose a hybrid multi-task residual network (MTRN), learning model, conduct prestack inversion. designed MTRN consists shared subnet mainly formed different units task-related subnets five fully connected...