散射波场的深度学习反演成像法  

Deep learning inversion imaging method for scattered wavefield

作　　者：奚先[1] 黄江清 XI Xian;HUANG Jiang-qing(China University of C,eosciences (Wuhan),School of Mathematics and Physics,Wuhan 430074,China;China University of Geoscienees (Wuhan),School Hospital,Wuhan 430074,China)

机构地区：[1]中国地质大学(武汉)数学与物理学院,武汉430074 [2]中国地质大学(武汉)校医院,武汉430074

出　　处：《地球物理学进展》2018年第6期2483-2489,共7页Progress in Geophysics

摘　　要：本文提出了一种散射波场的卷积神经网络深度学习反演成像方法.我们提出了三种散射距离场概念,由此成功地实现了三种卷积神经网络的深度学习训练及其反演.经过训练的三个CNN网络都可以应用于各种十分复杂的地震散射波场的反演,具有良好、稳健的反演能力和泛化能力且三种反演结果各具特色可以相互借鉴.将散射波场输入CNN网络后得到的输出(反演结果)图像非常直观容易辨识,可以大致辨识出测试模型中各散射点的准确位置,可以让一个不懂地震记录的外行从一个全新的视角去分析处理复杂的波场记录.A method of deep learning inversion imaging based on convolutional neural network for scattered wavefield is presented in this paper.We put forward three concepts of scattering distance field,Thus,deep learning training and inversion of three convolutional neural networks are successfully implemented.The three CNN networks trained by the training models can be applied to the inversion of various complex seismic scattered wave fields,With good and robust inversion ability and generalization ability,and the three kinds of inversion results have their own characteristics,they can be used for reference.The image of the output (inversion result)obtained after the input of the scattered wave field to the CNN network is very intuitive and easy to identify,can be identified accurately the position of each scattering point in the test model, can let a layman not understand seismic records from a new perspective to analyze the complex wave field record the.The output (scattering distance field)image obtained from the scattering wave field (test model)input the CNN network is very intuitive and easy to identify,The exact location of each scattering point in the model can be roughly identified,It allows a layman who does not understand seismic records to analyze complex wave field records from a new perspective.

关 键 词：散射距离场 深度学习 散射波反演 卷积神经网络成像 

分 类 号：P631[天文地球—地质矿产勘探]