基于MEGNet的岩心孔洞图像分割算法

打开文本图片集

中图分类号：TN911.73-34 文献标识码：A 文章编号：1004-373X（2026）09-0079-08

Abstract：Inviewof thehighsimilaritybetwenforegroundandbackground，largediferenceinholemorphologyand blurred edgesegmentationincoreholesegmentation，thispaperproposesanedge-guidedsegmentationmodelMEGNetbasedon Mamba.Firstly，theparalelPVMBlockisdesignedasthebackbonemodule tomodel thelong-rangedependence，whicheduces thenumberofparametersandaleviatesthemissegmentation.Secondly，anedgegeneration（EG）moduleisconstructedto generateedgefeaturesbyfusinglow-leveldetailsandhigh-levelsemanticfeatures.Then，andge-guidedatention（EGA）module is proposed tooptimizeedgedetailsbycombiningreversefeaturesandmulti-scalechannelatentionmodule（MS-CAM）.Finally， thefeatureenhancementmodule（FEM）isintroduced，andtheulti-scaleontextinformationiscapturedbymulti-expansionate dilated convolution to enhance the expression of key features.Experiments show that the F1 -score，intersection over union （IoU） andmeanintersectionoverunion （MIoU）of MEGNetonthecore holedatasetreach88.83%，79.92%and89.73%，respectively. The proposed methodhasbetersegmentationeffectandexcelent performance in comparison withthe mainstreamsemantic segmentation models.

Keywords：core hole;deep learning; Mamba; edge feature; feature enhancement;attention module

0 引言

岩心孔洞的结构是油气储层的重要特征，不仅能反映储层含油气的丰富度，还直接决定油气藏产能的差异分布，对油气勘测与开发具有重要意义，是油气勘探领域的研究重点。（剩余12881字）