Faster R-CNN在储层裂缝识别中的初探  

Application of Faster R-CNN in Reservoir Fractures Identification

作　　者：魏伯阳 杨占军 王立党 赵阳 李昂 WEI Boyang;YANG Zhanjun;WANG Lidang;ZHAO Yang;LI Ang(Henan Province Geological Survey Institute,Zhengzhou 450000,China)

机构地区：[1]河南省地质研究院,河南郑州450000

出　　处：《河南水利与南水北调》2024年第11期62-63,共2页Henan Water Resources & South-to-North Water Diversion

摘　　要：裂缝的检测和参数提取对储层来讲是重点也是难点。区域卷积神经网络(R-CNN)技术的发展给裂缝的计算机自动识别带来了新思路。R-CNN将分类问题转换为回归问题;Fast R-CNN将边框回归纳入训练过程,减少了边框回归所需要的时间;Faster R-CNN改变了输出建议区域的方法,大幅减少了输出建议区域消耗的时间。文章将Faster R-CNN在测井成像识别裂缝的初步应用效果优秀,检测得到裂缝区域及参数过程耗时0.041s,经验证Faster R-CNN是一种有效的可以识别裂缝并提取裂缝参数的方法。Fracture detection and parameter extraction is the key and difficult point for reservoir.The development of the regional convolution neural network(R-CNN)provides a new idea for computer automatic identification of fractures.The R-CNN converts classification problems into regression problems.The Fast R-CNN incorporates the per-class bounding-box regression into the training process,which could shorten the time spent in the per-class bounding-box regression.By changing the way to output the bounding-box,the Faster R-CNN could greatly shorten the time spent in the output bounding-box.In this paper,the preliminary application of Faster R-CNN in logging imaging identification of fractures presents excellent results.The time for detection of fracture area and parameters is 0.041s.It is verified that Faster R-CNN is an effective method to identify fractures and extract fracture parameters.

关 键 词：储层裂缝 自动识别 Faster R-CNN 

分 类 号：P631.81[天文地球—地质矿产勘探]