基于学习的地震剖面超分辨率重建方法研究  

Research on super resolution reconstruction method of seismic profile based on learning

作　　者：刘旭跃[1] 黄骏[1] LIU Xuyue;HUANG Jun(Sinopec Geophysical Research Institute,NanJing 211103,China)

机构地区：[1]中国石化石油物探技术研究院,南京211103

出　　处：《物探化探计算技术》2018年第1期20-26,共7页Computing Techniques For Geophysical and Geochemical Exploration

基　　金：中石化科技部项目(P14152)

摘　　要：地震信号在传输、采集、接收等过程中不可避免会受各种不利因素影响,为提高由信号生成的地震剖面图像质量,采用对单幅地震剖面图像进行处理,首先采用TV全变分方法,有效去除干扰波并保持图像纹理和边缘,然后从原始低分辨率图像中获得局部信息,运用逐层相似性学习方法进行高分辨率图像地重建,尤其在对图像进行放大时,仍能较清晰还原地震剖面图像纹理细节特征。经实际剖面测试表明,图像重建质量有所提升,在视觉效果、峰值信噪比等方面均获得了较好结果。As a result of the seismic signal is inevitably affected by various unfavorable factors in the process of transmission,acquisition and reception.In order to improve the quality of seismic images generated by signals,this paper deals with a single seismic image.Firstly,the TV full variational method is adopted to effectively remove the interference wave and keep the image texture and the edge,then the local information is obtained from the original low-resolution image,and then the reconstruction of the high-resolution image is carried out by using the layer-by-layer similarity learning method,especially when the image is enlarged,the texture detail characteristic of the seismic profile image can be reduced clearly.The actual profile test shows that the quality of the image reconstruction is improved,and the better results are obtained in the visual effect,the peak signal to noise ratio and so on.

关 键 词：地震剖面 局部信息 超分辨率 相似性学习 

分 类 号：P631.4[天文地球—地质矿产勘探]