基于循环生成对抗网络的逆时偏移成像结果优化  

Optimization of migrated images based on cycle generative adversarial network

作　　者：黄建平[1,2] 刘博文 黄韵博 孙加星 李亚林 雷刚林[4] 段文胜[4] 陈飞旭[1,4] 侯中根 HUANG Jianping;LIU Bowen;HUANG Yunbo;SUN Jiaxing;LI Yalin;LEI Ganglin;DUAN Wensheng;CHEN Feixu;HOU Zhonggen(School of Geosciences in China University of Petroleum(East China),Qingdao 266580,China;Pilot National Laboratory for Marine Science and Technology(Qingdao),Qingdao 266071,China;School of Pharmacy,Dezhou University,Dezhou 253004,China;Tarim Oilfield Branch,CNPC,Korla 841000,China;School of Chemistry and Chemical Engineering,University of Jinan,Jinan 250022,China)

机构地区：[1]中国石油大学(华东)地球科学与技术学院,山东青岛266580 [2]海洋国家实验室海洋矿产资源评价与探测技术功能实验室,山东青岛266071 [3]德州学院药学院,山东德州253004 [4]中国石油塔里木油田公司,新疆库尔勒841000 [5]济南大学化学化工学院,山东济南250022

出　　处：《中国石油大学学报（自然科学版）》2024年第5期36-45,共10页Journal of China University of Petroleum(Edition of Natural Science)

基　　金：“十四五”重大项目(2021QNLM020001);国家重点研发计划(2019YFC0605503C);中石油重大项目(ZD2019-183-003);优秀青年科学基金项目(41922028);国家创新群体项目(41821002)。

摘　　要：在常规逆时偏移方法基础上,通过引入循环生成对抗网络(CycleGAN)发展一种基于循环生成对抗网络的逆时偏移成像结果优化方法。首先构建包含两个生成器和两个判别器的CycleGAN。在对抗损失、循环一致性损失函数的基础上,添加身份损失函数,以避免训练过度;然后,组建样本集来训练网络,使其学习常规逆时偏移成像结果和最小二乘逆时偏移成像结果之间的映射关系;最后,利用其他合成数据和实际资料测试网络效果。结果表明,提出的基于循环生成对抗网络的逆时偏移成像结果优化方法在获得高精度、高信噪比成像结果的同时有效地提高了计算效率。This paper presents an optimization method for reverse time migration(RTM)imaging using a cycle generative adversarial network(CycleGAN).Based on the conventional RTM method,the CycleGAN framework is introduced,comprising two generators and two discriminators.To prevent overfitting,an identity loss function is added alongside the adversarial and cyclic consistency loss function.A training dataset is constructed to train the network,enabling it to learn the mapping between the conventional RTM imaging results and least-squares RTM imaging results.Finally,synthetic data and real data are used to test the network s performance.The prediction results from both datasets demonstrate that the proposed method effectively enhances computational efficiency while producing high-precision and high-SNR imaging results.

关 键 词：循环生成对抗网络 残差网络 逆Hessian 最小二乘逆时偏移 

分 类 号：P631.4[天文地球—地质矿产勘探]