摘要
针对地球物理勘探中,采集的地震数据被大量随机噪声所覆盖的问题,提出基于改进生成对抗网络(GAN)的地震数据随机噪声压制方法。该网络的核心是将卷积层引入GAN,建立合适的生成器与判别器并引入LeakyReLU、Sigmoid激活函数以优化网络训练效果;卷积层代替全连接层,通过局部连接与权值共享实现在保留数据有效信息的同时提升计算效率。以实际地震数据与合成地震数据进行实验,在数据可视化、峰值信噪比的评价指标下,结果表明与原始GAN、传统方法相比,该方法在不同噪声水平下均有较好的去噪效果,有利于后续地震资料解释等环节。
To solve the problem that the seismic data acquired in geophysical exploration are covered by a large number of random noises,a random noise suppression method of seismic data based on improved Generative Adversarial Network(GAN)is proposed.The core of the network is to introduce convolution layer into GAN,establish appropriate generator and discriminator,and introduce LeakyReLU and Sigmoid activation functions to optimize the network training effect.The convolution layer replaces the full connection layer,and improves the computing efficiency while retaining the effective information of data through local connection and weight sharing.The experiment is carried out with actual seismic data and synthetic seismic data.Under the evaluation indexes of data visualization and peak signal to noise ratio,the results show that compared with the original GAN and traditional methods,this method has better denoising effect under different noise levels,which is conducive to subsequent seismic data interpretation and other links.
作者
王林霖
王姣
王中训
刘培学
韩丽
殷振凯
WANG Lin-lin;WANG Jiao;WANG Zhong-xun;LIU Pei-xue;HAN Li;YIN Zhen-kai(Yantai University,Physics and Electronic Information College,Yantai 264003,Shandong Province,China;Qingdao Huanghai University,IM College,Qingdao 266000,Shandong Province,China)
出处
《信息技术》
2024年第3期10-15,共6页
Information Technology
基金
山东省自然科学基金青年基金(ZR2022QD112)
山东省重点研发计划(2019GGX105001)
青岛黄海学院博士科研启动基金项目(2020boshi01)
烟台大学研究生科研创新基金资助(GGIFYTU2316)。
关键词
生成对抗网络
卷积层
地震数据
随机噪声
去噪
Generative Adversarial Network(GAN)
convolution layer
seismic data
random noise
denoising
