Deep neural networks have enabled progress in a wide variety of applications. Growing the size network typically results improved accuracy. As model sizes grow, memory and compute requirements for training these models also increases. We introduce technique to train deep using half precision floating point numbers. In our technique, weights, activations gradients are stored IEEE half-precision format. Half-precision numbers limited numerical range compared single-precision propose two...

Seismic data reconstruction is one of the essential steps in seismic processing. Recently, deep learning (DL) models have attracted huge attention exploration, which has been applied to reconstruction, especially convolutional neural network (CNN)-based methods. However, general CNN-based only consider features time domain and do not take into account frequency features. Moreover, there are detailed lost due downsampling scheme. We propose a wavelet-based residual DL (WRDL) reconstruct...

The seismic horizon is a critical input for the structure and stratigraphy modeling of reservoirs. It extremely hard to automatically obtain an accurate interpretation data in which lateral continuity reflections interrupted by faults unconformities. process can be viewed as segmenting traces into different parts each part unique object. Thus, we have considered object detection problem. We use encoder-decoder convolutional neural network (CNN) detect “objects” contained traces. boundary...

Horizon picking is of paramount importance in seismic interpretation because it has a significant impact on subsequent and inversion. Although manual various automatic methods have been widely used for horizon picking, they still several problems, such as being time-consuming highly dependent human experience. Recently, deep learning implemented to solve these problems. However, traditional convolutional neural networks shortage capturing global features, Vision Transformers, recently...

Microseismic imaging plays an important role in hydraulic fracture detection, and the first-arrival picking of microseismic events is bedrock imaging. Manual most reliable also time-consuming method for detection first arrival events. Accurate efficient a real noisy environment challenge automatic methods. We have developed novel workflow to automatically pick microseismics by using state-of-the art pixel-wise convolutional image segmentation method. form training data randomly selecting...

Seismic noise attenuation is an important step in seismic data processing. Most algorithms are based on the analysis of time-frequency characteristics and noise. We have aimed to attenuate white using convolutional neural network (CNN). Traditional CNN-based need prior information (the “clean” or contained seismic) training process. However, it difficult obtain such practice. assume that can be simulated by a sufficient number user-generated realizations. then modified denoising CNN...

Abstract Horizontal tracking constitutes a foundational and critical aspect of seismic data processing interpretation. The traditional manual horizon work is time-consuming subjectively interferes with huge factors. In response, automated methods have been developed, albeit the drawback requiring extensive parameterization. Intelligent combined deep learning are also popular development direction in future. Nonetheless, efficacy conventional convolutional neural network (CNN) approaches,...

Abstract Seismic data reconstruction is an essential process to reduce the effects of missing traces in field acquisition. Imputation and interpolation techniques have been proposed reconstruct subtle features, which a challenging task that remains be solved, such as recovery capability for weak reflections consecutively parts computational efficiency. In recent years, deep learning (DL), especially residual network (ResNet), has gained remarkable success seismic because it can precisely...

To better reveal time-varying spectral components of nonstationary seismic signals, time-frequency analysis (TFA) has been widely applied in processing and analysis. In this letter, we propose an advanced TFA method based on optimal mode separation adaptive wavelet bank design. The proposed separation-based transform (AMSWT) generates a superior resolution. addition, because the is adaptively built intrinsic modes, ability to accurately characterize geophysical structures significantly...

Seismic volumetric dip and azimuth are widely used in assisting seismic interpretation to depict geologic structures such as chaotic slumps, fans, faults, unconformities. Current popular estimation methods include the semblance-based multiple window scanning (MWS) method gradient structure tensor (GST) analysis. However, accuracy using semblance is affected by of reflectors. The GST analysis centered at point. We have developed a new algorithm overcome disadvantages MWS combining improving...

The stratigraphic correlation of well logs is crucial for characterizing subsurface reservoirs. However, due to the complexity and huge amount data, manual time- resource-intensive. Hence, various computerized methods have been developed, especially regarding convolutional neural networks (CNNs). Recently, Transformer, a self-attention system that evolved from Natural Language Processing (NLP), has attained state-of-the-art performance over CNNs in variety domains because its ability...

Characterization of facies within a hydrocarbon reservoir is essential for potential prospect identification and evaluation. We have developed practical workflow that integrates poststack seismic attributes well-log analysis to understand the development depositional setting Triassic-age Akekule Formation in Tahe field, Northwest China. The begins with sequence sedimentary cycle on selected benchmark wells. then identify sand bodies each using well logs. from logs drilling cuttings together...

Seismic horizons are geologically significant surfaces that can be used for building geology structure and stratigraphy models. However, horizon tracking in 3D seismic data is a time-consuming challenging problem. Relief human from the tedious interpretation one of hot research topics. We proposed novel automatically method by using deep convolutional neural network. employ state-of-art end-to-end semantic segmentation to track automatically. Experiment result shows our network multiple...

Long Short-Term Memory Recurrent neural networks are generally used in speech recognition, machine translation and other fields. And LSTM-RNN also performs well data anomaly detection. However, Due to the repeatability of LSTM-RNN, general-purpose processors such as CPU GPGPU cannot efficiently implement most existing model optimizations on FPGA aimed at LSTM cells or large-scale accelerations that do not require high accuracy (such recognition). For aircraft detection, which models with...

The seismic quality factor [Formula: see text] quantifies the anelastic attenuation of waves in subsurface and can be used assisting reservoir characterization as an indicator hydrocarbons. Usually, text]-factor is estimated by comparing spectrum changes vertical profiles poststack data. However, processing such normal moveout (NMO) stretch would distort Hence, we have using prestack time migration gathers. To mitigate NMO effect, compensate gathers time-frequency domain. Similar to log...

Microseismic event picking is one of the key steps in seismic processing and imaging. Manually a widely used way to pick microseismic events, which time-consuming. The standard short-term average/long-term average (STA/LTA) traditional method first arrivals, would lead inaccurate first-arrival picks case low signal-to-noise ratio (SNR). We developed workflow automatically arrivals by using feature pyramid networks (FPNs). To train proposed model, we randomly select part traces manually time...

Abstract Visual prostheses offer a possibility of restoring vision to the blind. It is necessary determine minimum requirements for daily visual tasks. To investigate recognition common objects in life based on simulated irregular phosphene maps, effect four parameters (resolution, distortion, dropout percentage, and gray scale) object was investigated. The results showed that accuracy significantly increased with an increase resolution. Distortion percentage had significant impact...