摘要
为解决浅地层剖面数据噪声多、分辨率低问题,本文将环形生成对抗网络的方法应用于浅地层剖面资料的去噪,实现智能去噪。首先,选择具有特殊对称生成对抗网络循环机制以及“循环一致性”损失的环形生成对抗性网络,并对其进行结构改进,提升网络学习和训练的性能。然后,基于优化的浅地层剖面样本集训练网络,实现对于浅地层剖面数据随机噪声的去除,提升数据的信噪比。通过对实验和实际资料的试算,以及与传统带通滤波方法的对比,验证本文方法对浅地层剖面数据去噪的有效性和适应性。
This study applied the cycle-consistent generative adversarial network method to the denoising of shallow profile data to realize intelligent denoising. This could help resolve the problem of noise and low resolution of shallow profile data. To do this, the cycle generative adversarial network with special symmetric generation countermeasure network cycle mechanism and "cycle consistency loss" was selected. We improved the performance of the network learning and training by optimizing the network structure. Next, based on the optimized shallow profile sample set training network, random noise was removed from the shallow profile data and the signal-to-noise ratio of the data was improved. The effectiveness and adaptability of this method for denoising shallow profile data were verified by trial calculations of experimental and actual data and by comparison with the traditional band-pass filtering method.
作者
张一
丁仁伟
赵硕
孙世民
韩天娇
ZHANG Yi;DING Renwei;ZHAO Shuo;SUN Shimin;HAN Tianjiao(College of Earth Science and Engineering,Shandong University of Science and Technology,Qingdao 266590,China)
出处
《CT理论与应用研究（中英文）》
2023年第1期15-25,共11页
Computerized Tomography Theory and Applications
关键词
人工智能
浅地层剖面
随机噪声
数据去噪
artificial intelligence
shallow profile data
random noise
data denoising
作者简介
张一,女,山东科技大学地球科学与工程学院硕士研究生,主要从事人工智能地球物理应用研究,E-mail:minniez423@163.com;丁仁伟,男,博士,山东科技大学地球科学与工程学院讲师,主要从事地球信息科学与技术、高性能计算、地震成像与速度建模理论与方法、地震数据智能处理与解释等方面研究,E-mail:dingrenwei@126.com。
