1 CONSS Contrastive Learning Approach for Semi-Supervised Seismic Facies Classiﬁcation

2025-04-28 0 0 2.46MB 5 页 10玖币

侵权投诉

CONSS: Contrastive Learning Approach for

Semi-Supervised Seismic Facies Classiﬁcation

Kewen Li, Wenlong Liu, Yimin Dou, Zhifeng Xu, Hongjie Duan, Ruilin Jing

Abstract—Recently, seismic facies classiﬁcation based on con-

volutional neural networks (CNN) has garnered signiﬁcant re-

search interest. However, existing CNN-based supervised learning

approaches necessitate massive labeled data. Labeling is laborious

and time-consuming, particularly for 3D seismic data volumes.

To overcome this challenge, we propose a semi-supervised method

based on pixel-level contrastive learning, termed CONSS, which

can efﬁciently identify seismic facies using only 1% of the

original annotations. Furthermore, the absence of a uniﬁed data

division and standardized metrics hinders the fair comparison of

various facies classiﬁcation approaches. To this end, we develop

an objective benchmark for the evaluation of semi-supervised

methods, including self-training, consistency regularization, and

the proposed CONSS. Our benchmark is publicly available to

enable researchers to objectively compare different approaches.

Experimental results demonstrate that our approach achieves

state-of-the-art performance on the F3 survey. Our all codes and

data are available at https://github.com/upcliuwenlong/CONSS.

Index Terms—Seismic Facies Classiﬁcation, Semi-Supervised

Learning, Seismic Interpretation, Contrastive Learning.

I. INTRODUCTION

SEISMIC facies classiﬁcation refers to the interpretation

of facies type from the seismic reﬂector information. It

is an important ﬁrst step in exploration, prospecting, reservoir

characterization, and ﬁeld development. Data from core and

well log offer a vertical perspective that can aid in the

interpretation of seismic facies. However, due to the expensive

cost of drilling operation, direct facies information is scarce.

Alternatively, manual assignment of seismic facies based on

seismic attributes is possible, yet remains a highly subjective

process that relies heavily on the experience of the seismic

interpreter.

Deep learning has garnered signiﬁcant popularity for seis-

mic data processing and interpretation,suah as denoising [1],

inversion [2], interpolation [3] ,fault detection [4] [5], etc.

The powerful feature extraction and representation capabilities

of neural networks enable deep learning methods to mitigate

human subjectivity. Zhao et al. [6] reviewed seismic facies

classiﬁcation and reservoir prediction approaches based on

deep convolutional neural networks. Alaudah et al. [7] pro-

posed a facies classiﬁcation framework based on deconvo-

lution network. These supervised deep learning methods are

The corresponding author is Kewen Li. likw@upc.edu.cn

Kewen Li, Wenlong Liu, Yimin Dou, Zhifeng Xu, College of computer

science and technology, China University of Petroleum (East China) Qingdao,

China.

Hongjie Duan, Ruilin Jing, Shengli Oilﬁeld Company, SINOPEC Dongying,

China.

This work was supported by grants from the National Natural Science

Foundation of China (Major Program, No.51991365),and the Natural Science

Foundation of Shandong Province, China (No.ZR2021MF082).

Fig. 1. Comparison of the fully supervised method with the proposed CONSS.

Fully supervised methods utilize labeled seismic facies data. CONSS extends

the contrastive learning branch to learning unlabeled data features.

dependent on labeled data and are incapable of acquiring

knowledge from unlabeled data. However, the labeling of 3D

seismic data volumes is a demanding and time-intensive task,

often requiring geological teams to devote hundreds of hours.

In the majority of cases, there exists a scarcity of labeled

data with an abundance of unlabeled data. Training with

unlabeled data will minimize the labeling cost, which is

a primary objective of semi-supervised learning. Saleem et

al. [8] implemented a semi-supervised approach based on

self-training. They use a model trained on labeled seismic

data to predict the label of unlabeled seismic data, with the

predicted label subsequently added to the training set as a

pseudo-label for retraining. Self-training is categorized as an

ofﬂine learning method and is susceptible to the inﬂuence

of noise pseudo-labels, leading to potential performance sat-

uration. Consistency regularization represents another semi-

supervised paradigm, which operates under the foundational

assumption that slight perturbations ought not to produce

signiﬁcant changes in the model’s output. In our experiments,

we implement the cross pseudo supervision [9] (CPS), which

yielded superior outcomes when compared to self-training

in the F3 survey. However, CPS necessitates the alternative

training of two models, thereby resulting in a comparatively

higher training overhead.

Contrastive learning [10] [11] [12] is self-supervised learn-

ing in which a neural network is trained to identify the

similarities and differences between different inputs. The main

idea of contrastive learning is to learn a representation that

can distinguish between positive and negative sample pairs.

In contrastive learning, a positive pair is a pair of samples

that are similar to each other, while a negative pair is a

pair of samples that are dissimilar. The network is trained to

arXiv:2210.04776v3 [cs.CV] 12 Mar 2023

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

1CONSS:ContrastiveLearningApproachforSemi-SupervisedSeismicFaciesClassicationKewenLi,WenlongLiu,YiminDou,ZhifengXu,HongjieDuan,RuilinJingAbstractRecently,seismicfaciesclassicationbasedoncon-volutionalneuralnetworks(CNN)hasgarneredsignicantre-searchinterest.However,existingCNN-basedsupervisedlear...

展开>> 收起<<

1 CONSS Contrastive Learning Approach for Semi-Supervised Seismic Facies Classiﬁcation.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

1 CONSS Contrastive Learning Approach for Semi-Supervised Seismic Facies Classiﬁcation

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: