基于UNet结构生成对抗网络的海底地震勘探数据混叠噪声压制方法  

Ocean Bottom Seismic Based on Generation of Countermeasure Network of Unet Structure Data Aliasing Noise Suppression Method

作　　者：童思友[1,2] 刘岗 徐秀刚[1,2] 王忠成 王金刚 杨德宽 Tong Siyou;Liu Gang;Xu Xiugang;Wang Zhongcheng;Wang Jingang;Yang Dekuan(College of Marine Geosciences,Ocean University of China,Qingdao 266100,China;Key Laboratory of Submarine Science and Exploration Technology,Ministry of Education,Ocean University of China,Qingdao 266100,China;Geophysical Exploration Research Institute of Sinopec Shengli Oilfield Branch,Dongying 257022,China;Shengli Branch of Sinopec Geophysical Company,Dongying 257022,China)

机构地区：[1]中国海洋大学海洋地球科学学院,山东青岛266100 [2]中国海洋大学海底科学与探测技术教育部重点实验室,山东青岛266100 [3]中国石化胜利油田分公司物探研究院,山东东营257022 [4]中国石化地球物理公司胜利分公司,山东东营257022

出　　处：《中国海洋大学学报（自然科学版）》2024年第8期123-131,共9页Periodical of Ocean University of China

基　　金：国家自然科学基金项目(42074140)资助。

摘　　要：本文提出一种基于UNet结构生成对抗网络(Pix2PixGAN)的海底地震勘探数据混叠噪声压制方法,该神经网络主要在于构建了一个适用于混叠噪声压制的生成器和判别器,生成器是UNet结构,可以提取和融合数据的特征映射信息,而通过加入skip-connection,可以保留更多的细节信息;判别器由两个卷积模块构成,通过PatchGAN输出多个固定大小的数据体,加入Dropout2d层,优化判别器的训练速度。通过制作的四千余个数据集对网络模型进行训练,将得到的训练参数加载到测试网络中,经过测试数据集的验证以及和常规的去噪方法相比,验证了本文采用的混叠噪声压制方法具有较高的压制精度和效率。The authors of proposes an aliasing noise suppression method for seabed seismic data based on UNet structure generation Counteraction network(Pix2PixGAN).The neural network mainly constructs a generator and discriminator suitable for aliasing noise suppression.The generator is an UNet structure,which can extract and integrate feature mapping information of data.By adding skip-connection,more details can be preserved;The discriminator is composed of two convolutional modules.Through PatchGAN,multiple data volumes of fixed size are output and Dropout2d layer is added to optimize the training speed of the discriminator.More than four thousand data sets were made to train the network model,and the obtained training parameters were loaded into the test network.After the verification of the test data sets and compared with the conventional denoising methods,it is verified that the aliasing noise suppression method adopted in this paper has higher suppression accuracy and efficiency.

关 键 词：生成对抗网络 海底地震勘探 地震数据 混叠 噪声压制 

分 类 号：P736[天文地球—海洋地质]