Most existing image inpainting methods aim to fill in the missing content inside-hole region of target image. However, areas be restored realistically degraded images are unspecified. Previous studies have failed recover degradations due absence explicit mask indication. Meanwhile, inconsistent patterns blended complexly with content. Therefore, estimating whether certain pixels out distribution and considering object is consistent context necessary. Motivated by these observations, a...

Depicting faults in seismic data is one of the key steps structure interpretation. However, manual identification a time-consuming and tedious process. In conventional methods, attributes associated with reflection continuities or discontinuities are extracted for fault detection. recent years, convolutional neural networks (CNNs) have been introduced to solve geophysical problems. Herein, we proposed powerful efficient methods enhance performance CNN-based We first introduce human reasoning...

Unsupervised image segmentation is an essential topic in the field of computer vision, which broke limitation training data and expanded application scenarios. Off-the-shelf clustering methods simply rely on semantic concepts incomplete boundary cues, resulting incorrect object boundaries. Therefore, this paper proposes unsupervised framework combining differentiable double (DDC) edge-aware superpixel (EA), outperform prior work accuracy art. First, a multi-layer feature extraction network...

Most existing arbitrary shape text detection methods employ connected components and center lines for grouping instances, which assume that texts in adjacent positions belong to the same instance. However, many hard-to-group scene are too complex be effectively processed this way. To address challenge, we propose a novel text-spotting method utilizes feature-based clustering inspired by human cognitive principles of perception. Our approach involves first utilizing character spotter obtain...

The offset vector tile (OVT) is a special prestack seismic gather type, and OVT technology processing suitable for wide-azimuth (WAZ) exploration. Because current WAZ interpretation not yet mature because of the lack key techniques tools OVT-domain interpretation, abundant azimuth information possessed in gathers has been fully used. To take full advantage to achieve more accurate geologic, reservoir, fluid information, we have developed an typical workflow. composed five analysis focused on...

Abstract Faults generated by seismic motion and stratigraphic lithology changes are essential research objects for hydrocarbon prospecting. This paper emphatically concentrates on the fault reconstruction from existing probability volume. The core idea is to transform separation of different sticks into a fitting segmentation problem point cloud data. First, we utilize filtering algorithm preprocess volume then complete coarse region growth algorithm. For intersecting faults, employ an...

Paleokarst systems, found in carbonate rock formations worldwide, have potential for creating vast reservoirs and facilitating hydrocarbon migration. Thus, studying these systems is essential the exploration development of reservoirs. An approach using convolutional neural networks (CNNs) introduced to automatically precisely identify cave features within 3D seismic data. efficient technique outlined generating ample amounts training data, which comprised synthetic data labels contained as a...

Hydrocarbon detection technology based on double‐phase medium theory was developed in 2004, and has been applied to hundreds of known wells many fields recent years. The agreement between results well data is over 80%. For 15 unknown wells, 13 among which showed consistent with that delivered by hydrocarbon detection, the conformability up 87%. Taking an example from a high‐production C4 block located Huabei Oil Field, this paper demonstrates application theory. First basic concepts are...

PreviousNext No AccessSEG 2017 Workshop: Carbonate Reservoir E&P Workshop, Chengdu, China, 22-24 October 2017The interpretative processing technology and its application of small-scale fracture-cave predictionAuthors: Xiangwen Li*Shifan ZhanYonglei LiuLei LiChunfeng TaoXiangwen Li*BGP Int., CNPC, 072751, Shifan ZhanBGP Yonglei LiuBGP Lei LiBGP Chunfeng TaoBGP 072751https://doi.org/10.1190/carbonate2017-36 SectionsAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions...

Summary With the development of wide-azimuth and high-density seismic exploration, massive full-azimuthal pre-stack data can be obtained, which contain abundant geological information. These not only 3D coordinate but also offset azimuth, named five-dimensional (3D + azimuth) data. However, how to manage, analyze display especially OVT gather, fully excavate fluid information contained in data, is main problem constraining interpretation. Based on characteristics we firstly carried out...

Different from purely data-driven supervised deep learning, we propose a theory-guided model to autonomously produce horizon volumes by calculating relative geologic time (RGT). To enhance model's generalization capability, integrate two unsupervised losses into the network, drawing upon domain knowledge of traditional automatic tracking. During optimization process, use manual interpretation results as data constraints regulate volume generation. Ultimately, this leads establishment...

One of the major challenges in deep learning inversion is insufficient number well logs, which leads to a scarcity labels that adequately capture vertical lithology patterns and lateral variations within geological formations. Forward modeling after log interpolation one commonly used approach augment labels. However, it crucial ascertain rationale behind such determine optimal method enhance accuracy. Through model testing, we confirm Deep Learning (DL) networks have ability tolerate...

In this work, we propose a weakly supervised learning method which could utilize sparse manual interpretation results as training data for 3D fault detection task. Following setting, design the masked data, are gathered from field seismic volumes, and loss function process. Synthetic volumes applied to testify proposed method. While make no claim that these better than predicted by full methods, believe can provide at least competitive with models in literature highlight potential of framework.