吕尧^1,2,3,,
单小彩^1,2,3,
霍守东^1,2,
杨长春^1,2,,
1. 中国科学院地质与地球物理研究所, 中国科学院油气资源研究重点实验室, 北京 100029
2. 中国科学院地球科学研究院, 北京 100029
3. 中国科学院大学, 北京 100049

基金项目: 中国科学院A类战略性先导科技专项（XDA14040101, XDA14040300）, 中央高校基本科研业务费专项资金（ZY1927）资助


详细信息

作者简介: 吕尧, 男, 1990年生, 博士研究生, 研究方向为油储地球物理、机器学习.E-mail:lvyao@mail.iggcas.ac.cn

通讯作者: 杨长春, 男, 中国科学院地质与地球物理研究所研究员.E-mail:ccy@mail.iggcas.ac.cn

中图分类号: P631

收稿日期:2019-02-11
修回日期:2019-07-12
上线日期:2020-01-05

Local SNR estimation of seismic data based on deep convolutional neural network

Lü Yao^1,2,3,,
SHAN XiaoCai^1,2,3,
HUO ShouDong^1,2,
YANG ChangChun^1,2,,
1. Institute of Geology and Geophysics, Chinese Academy of Sciences, Key Laboratory of Petroleum Resources Research, CAS, Beijing 100029, China
2. Innovation Academy for Earth Science, Chinese Academy of Sciences, Beijing 100029, China
3. University of Chinese Academy of Sciences, Beijing 100049, China


More Information

Corresponding author: YANG ChangChun,E-mail:ccy@mail.iggcas.ac.cn

MSC: P631

--> Received Date: 11 February 2019
Revised Date: 12 July 2019
Available Online: 05 January 2020


摘要
摘要:估计地震数据的信噪比对于地震数据的处理和解释具有重要作用.以往估计地震数据信噪比的方法都需要分离数据中的有效信号和噪声, 然后再估计相应的信噪比.这些估计方法的精度严重依赖信号估计方法或噪声压制方法的有效性, 往往存在偏差.本文提出一种估计地震数据局部信噪比的深度卷积神经网络模型, 通过迭代训练优化参数, 构建从含噪地震数据到其信噪比的特征映射.然后使用该神经网络完成信噪比的推理预测, 不需要分离地震数据中的有效信号和噪声.模拟数据和实际资料的处理结果都表明, 本文的方法可以准确而高效地估计局部地震数据的信噪比, 为地震数据质量的定量评价提供依据.
关键词: 局部信噪比估计/
深度卷积神经网络/
质量评价

Abstract:Signal-to-noise ratio (SNR) estimation of seismic data is important for the following data processing and interpretation. Almost all of the current algorithms for SNR estimation need to estimate the effective signals in the noisy data during calculations. However, there are always bias errors exiting in these SNR estimation results because the effective signals in the noisy data cannot be estimated accurately. In this paper, we propose a deep convolutional neural network architecture to estimate local SNR of the seismic data. Firstly, the network learns a feature mapping from a normalized noisy data patch to its SNR value by training and optimizing parameters. Then, the network can predict the SNR when it is fed in a seismic data patch, without the need of effective signals. Synthetic data and real data tests show the effectiveness and efficiency of our method to estimate the SNR and evaluate the quality of seismic data.
Key words:Local SNR estimation/
Deep convolutional neural network/
Quality evaluation



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=dqwlxb_15308