基于Gabor纹理学习的地震数据重建算法  

A texture feature learning method based on Gabor transform for seismic data interpolation

作　　者：贾永娜 吴杰 王国伟 田玲玲 顾军华 JIA Yongna;WU Jie;WANG Guowei;TIAN Lingling;GU Junhua(School of Artificial Intelligence,Hebei University of Technology,Tianjin300401,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300401

出　　处：《石油地球物理勘探》2023年第3期617-625,共9页Oil Geophysical Prospecting

基　　金：国家自然科学基金项目“基于特征张量的高维地震数据插值方法研究”(41804118);河北省自然科学基金项目“基于深层抽象结构特征的高维地震数据插值”(D2022202006)联合资助。

摘　　要：地震数据重建技术主要用于解决地震数据采集过程中出现的地震道缺失或空间采样不足等问题。针对支持向量回归(SVR)地震数据重建方法未充分利用地震数据物理信息的不足,引入可提取不同尺度和方向纹理特征的Gabor变换,旨在充分挖掘地震数据的物理信息,并基于SVR算法框架重建缺失的地震数据。即首先利用Gabor滤波器提取地震数据的纹理特征,并与原始数据特征相结合,构建新的特征向量数据库;然后通过SVR算法学习回归模型,用于重建缺失的地震数据。通过大量合成地震数据与实际地震数据重建实例,表明由Gabor变换提取的纹理特征能有效提高SVR算法的重建精度,并获得更高信噪比。The seismic data reconstruction technology is mainly employed to address the problems such as missing seismic traces or insufficient spatial sampling.As a seismic data reconstruction method,support vector regression does not make full use of the physical information of seismic data.Thus,this paper introduces Gabor transform which can extract texture features with different scales and directions.The purpose is to fully mine the physical information of seismic data and reconstruct missing seismic data based on the algorithm framework of support vector regression.Firstly,the texture features are extracted by the Gabor filters,and a new feature vector database can be built by combining with the original data features.Then,a regression model is learned by the support vector regression algorithm to reconstruct the missing seismic data.A large number of reconstruction examples from synthetic and actual seismic data show that texture features extracted by Gabor transform can effectively improve the reconstruction accuracy of support vector regression algorithms and realize higher signal-to-noise ratios.

关 键 词：地震数据重建 支持向量回归(SVR) 纹理特征 GABOR变换 

分 类 号：P631[天文地球—地质矿产勘探]