CUNet断层智能识别方法及其在断层识别中的应用研究  

CUNet deep learning method and its application in fault recognition

作　　者：陈继兴 路鹏飞 袁兆林 郭爱华 王丹丹[4] 刘春飞 CHEN JiXing;LU PengFei;YUAN ZhaoLin;GUO AiHua;WANG DanDan;LIU ChunFei(School of Information Engineering,East China University of Technology,Nanchang 330013,China;Jiangxi Engineering Laboratory of Radiofrequency Big Data Technology,East China University of Technology,Nanchang 330013,China;School of Earth Sciences,East China University of Technology,Nanchang 330013,China;Geological Research Institute of the First Oil Production Plant of North China Oilfield,Renqiu 062550,China)

机构地区：[1]东华理工大学信息工程学院,南昌330013 [2]东华理工大学,江西省放射性地学大数据技术工程实验室,南昌330013 [3]东华理工大学地球科学学院,南昌330013 [4]华北油田第一采油厂地质研究所,任丘062550

出　　处：《地球物理学进展》2024年第2期561-571,共11页Progress in Geophysics

基　　金：江西省核地学数据科学与系统工程技术研究中心开放基金项目“基于VNet深度学习架构的低序级断层智能识别方法研究”(JETRCNGDSS202205);“地震大数据高分辨率处理方法研究”(JETRCNGDSS202106);东华理工大学校级课题“地质大数据识别低序级断层方法研究”(DHBK2019222)联合资助。

摘　　要：断层智能识别在地震资料精细解释中起到了非常重要的作用,但目前,断层智能识别方法还存在着模型训练难度大,特征图像提取信号细节受损等问题.针对这些问题,本文基于UNet网络结构的基础上结合VNet深度学习网络的思想,提出了一种改进的CUNet(Convolution Unity Networking)断层智能识别方法.CUNet断层识别方法的网络结构是在UNet结构的基础上,利用卷积操作代替下采样中的最大池化,以及3D反置卷积操作代替上采样卷积操作,从而使CUNet结构可以增加信号的感受野,缓解信号细节特征在上采样和下采样操作中的损失,保留信号细节信息,增强图像特征的提取.实验结果表明CUNet网络结构在精度上达到了94.3%,与UNet网络结构相比提高了1.4%,同时起到了更好的抗过拟合效果.根据断层地震地质特征,将CUNet网络结构应用于地震图像的断层智能识别中.应用结果表明,该网络结构不仅更准确的检测到了断层特征,对断层分布刻画更细致,而且极大地节省了计算时间.Intelligent fault recognition plays a very important role in the fine interpretation of seismic data.However,at present,the intelligent fault recognition method still has problems such as difficulty in model training and damaged signal details in feature image extraction.In view of these problems,this paper proposes an improved CUNet(Convolution Unity Networking)fault intelligent recognition method based on the UNet network structure and the idea of VNet deep learning network.The CUNet fault identification method employs convolution operations to replace the max-pooling in downsampling and 3D transpose convolution operations to replace the convolution in upsampling,thus enhancing the receptive field of the CUNet structure and alleviating the signal detail loss during upsampling and downsampling operations.The experimental results show that the accuracy of CUNet network structure reaches 94.3%,which is 1.4%higher than that of UNet network structure,and has a better anti-over-fitting effect.According to the seismic geological characteristics of faults,the CUNet network structure is applied to the intelligent fault identification of seismic images.The application results show that the network structure not only detects the fault features more accurately,but also depicts the fault distribution more carefully,and greatly saves the calculation time.

关 键 词：断层智能识别 深度学习 CUNet网络 抗过拟合 断层特征 

分 类 号：P631[天文地球—地质矿产勘探]