基于局部特征匹配的井下管柱图像智能拼接融合技术  

Intelligent image mosaic and fusion technology of downhole pipe string based on local feature matching

作　　者：汤清源 杜宇成 叶胜 房伟 梁建龙 袁翔 刘浩浩 TANG Qingyuan;DU Yucheng;YE Sheng;FANG Wei;LIANG Jianlong;YUAN Xiang;LIU Haohao(Jianghan Machinery Research Institute Limited Company of CNPC,Jingzhou,Hubei 434099,China;School of Electrical and Electronic Engineering,Wuhan Polytechnic University,Wuhan,Hubei 430023,China;Downhole Service Company,PetroChina Chuanqing Drilling Engineering Co.,Ltd.,Chengdu,Sichuan 610000,China)

机构地区：[1]中石油江汉机械研究所有限公司,434001 [2]武汉轻工大学电气与电子工程学院,430048 [3]中国石油川庆钻探工程有限公司井下作业公司

出　　处：《天然气工业》2024年第9期190-198,共9页Natural Gas Industry

基　　金：中国石油天然气集团有限公司科学研究与技术开发项目“直属院所基础研究和战略储备技术研究”(编号:2021DQ03-02);中石油江汉机械研究所有限公司科学研究与技术开发项目“井下可视化定量分析系统研究”(编号:2022JJYSJ004)。

摘　　要：井下电视成像测井可以直观地监测井下管柱是否异常,但采集的井下管柱图像存在纹理低、光照不足、背景重复等问题,传统的尺度不变特征变换(Scale-Invariant Feature Transform,SIFT)等算法很难稳定地检测出高质量的特征点,导致图像拼接融合鲁棒性差。为此,基于局部特征匹配的思路,先利用反向像素映射算法将管柱图像展开成平面图,并对径向误差进行精确修正,再利用卷积神经网络提取局部特征,利用注意力机制在粗略层面上建立像素级匹配,最后引入最佳拼接线和平滑函数来消除拼接误差,实现了井下管柱大尺度图像的智能拼接融合。研究结果表明:(1)基于局部特征匹配的井下管柱图像智能拼接融合技术,通过图像预处理、特征匹配和图像融合,解决了井下管柱图像拼接融合的稳定性问题;(2)图像智能融合质量的平滑权重因子(k)为0.05时融合效果最佳,k值越小融合图像拼接缝越明显,k值过大则容易在重叠区域产生重影;(3)通过计算待拼接图像的最佳拼接线来消除角度倾斜带来的误差,达到了稳定智能拼接融合的目的;(4)与SIFT算法相比,该算法能检测出的特征点数量平均增加了74.6%,平均智能匹配正确率由83.9%增加到了98.8%。结论认为,该算法检测到的特征点数量和正确率都得到了明显提升,智能融合图像的结构相似性、峰值信噪比和均方误差等指标均优于传统算法,为解决井下管柱探测难题提供了新思路和技术手段。Downhole TV imaging logging can directly monitor the anomalies of downhole string,but the acquired downhole string images have problems such as low texture,insufficient illumination and repetitive background.Traditional algorithms such as scale-invariant feature transform(SIFT)can hardly detect high-quality feature points stably,resulting in poor robustness of image mosaic and fusion.In this paper,according to the idea of local feature matching,inverse pixel mapping algorithm is used to unfold the string image into a planar diagram,and the radial error is precisely corrected.Then,the convolutional neural network is used to extract local features,and the attention mechanism is adopted to establish pixel-level matching at the coarse level.Finally,the optimal mosaic line and the smoothing function are introduced to eliminate the mosaic error,and thus the intelligent mosaic and fusion of the large-scale image of downhole string is realized.And the following research results are obtained.First,the intelligent image mosaic and fusion technology of downhole pipe string based on local feature matching achieves a stable mosaic and fusion of downhole string image through image preprocessing,feature matching and image fusion.Second,when the smoothing weight factor(k)indicating the intelligent image fusion quality is 0.05,the fusion effect is the best.The smaller the k value,the more obvious the image mosaic seam.If the k value is too high,double image can be formed easily in the overlap zone.Third,the error caused by angle tilt is eliminated by calculating the optimal mosaic line of the to-be-treated image,so as to realize stable intelligent mosaic and fusion.Fourth,compared with the SIFT algorithm,the number of feature points that can be detected by this method is increased by 74.6%on average,and the average accuracy rate of intelligent matching is increased from 83.9%to 98.8%.In conclusion,the number and accuracy rate of detected feature points by this method are significantly improved,and the structural similarity,peak s

关 键 词：井下管柱图像 局部特征匹配 特征点 智能图像拼接 图像融合 图像预处理 卷积神经网络 结构相似性 

分 类 号：TE358[石油与天然气工程—油气田开发工程]