基于深度学习UNet++网络的初至波拾取方法  

UNet++:a deep-neural-network-based seismic arrival time picking method

作　　者：李薇薇[1] 龚仁彬[1] 周相广[1] 林霞[1] 米兰 李宁[1] 王晓东[2] 肖高杰[1] LI WeiWei;GONG RenBin;ZHOU XiangGuang;LIN Xia;MI Lan;LI Ning;WANG XiaoDong;XIAO GaoJie(Research Institute of Petroleum Exploration and Development,PetroChina,Beijing 100083,China;Bureau of Geophysical Prospecting INC,China National Petroleum Corporation,Renqiu 062552,China)

机构地区：[1]中国石油勘探开发研究院,北京100083 [2]中国石油集团东方地球物理勘探有限责任公司,任丘062552

出　　处：《地球物理学进展》2021年第1期187-194,共8页Progress in Geophysics

基　　金：中国石油认知计算平台试点项目(E8)资助。

摘　　要：初至波拾取是地震资料处理中一项基础而重要的工作.为解决我国西部沙漠、黄土塬、戈壁等地区地震资料信噪比低,致使初至波拾取准确率不高的难题.本文创新提出一种基于图像分割技术——UNet++神经网络应用于初至波智能拾取.输入原始地震数据及少量初至时间的标签数据进行监督学习,并建立UNet++模型,应用西部某工区地震数据测试,实验证明,UNet++模型性能稳定,炸药震源初至波拾取准确率达到98%,可控震源初至波拾取准确率达到98%.此外,本方法与商业软件、U-net网络的初至拾取对比表明,UNet++优势明显,具有准确率高,抗噪能力强,性能稳定、高效等特点.The first arrival picking is fundamental to seismic processing, especially for static correction. As irregular topography and random noises, it is an extremely difficult to breaks pick-up for seismic data with low signal-to-noise ratio in in piedmont of the southern Junggar Basin. We present a deep-neural-network based arrival-time picking method called UNet++.it uses seismic data as input and generates probability distributions of first arrivals as output. UNet++ is trained on the limited available data set provided by analyst-labeled first arrival times from field seismic data. UNet++ can effectively detect the position of first-break in a low SNR seismic data. The accuracy of UNet++ model achieved 98%, that demonstrate effectiveness and superiority of our method. While further testing against existing methods is required, we compare UNet++ with existing methods under identical experimental conditions, like commercial software,U-net.UNet++ achieves significant improvements for low SNR seismic data.The experimental have shown that UNet++ achieves much higher picking accuracy, and has a significant higher effectiveness than existing methods.

关 键 词：初至波 深度学习 UNet++ 

分 类 号：P631[天文地球—地质矿产勘探]