基于视觉大模型的砂岩岩石结构智能评价方法  

Intelligent evaluation of sandstone rock structure based on a visual large model

作　　者：任义丽 曾昌民[4,5] 李欣 刘茜[1,2] 胡延旭 苏乾潇 王孝明 林志威 周屹霄 郑紫路 胡蕙滢 杨艳宁 惠芳 REN Yili;ZENG Changmin;LI Xin;LIU Xi;HU Yanxu;SU Qianxiao;WANG Xiaoming;LIN Zhiwei;ZHOU Yixiao;ZHENG Zilu;HU Huiying;YANG Yanning;HUI Fang(PetroChina Research Institute of Petroleum Exploration&Development,Beijing 100083,China;Artificial Intelligence Technology R&D Center of Exploration and Development,CNPC,Beijing 100083,China;National Key Laboratory for Multi-resources Collaborative Green Production of Continental Shale Oil,Daqing 163000,China;PetroChina Tarim Oilfield Company,Korla 841000,China;Xinjiang Key Laboratory of Ultra-deep Oil and Gas,Korla 841000,China;Research Institute of Petroleum Exploration and Development,PetroChina Huabei Oilfield Company,Renqiu 062552,China;Peking University,Beijing 100871,China;No.6 Oil Production Plant,PetroChina Changqing Oilfield Company,Xi’an 710200,China)

机构地区：[1]中国石油勘探开发研究院,北京100083 [2]中国石油天然气集团有限公司勘探开发人工智能技术研发中心,北京100083 [3]多资源协同陆相页岩油绿色开采全国重点实验室,黑龙江大庆163000 [4]中国石油塔里木油田公司,新疆库尔勒841000 [5]新疆超深油气重点实验室,新疆库尔勒841000 [6]中国石油华北油田公司石油勘探开发研究院,河北任丘062552 [7]北京大学,北京100871 [8]中国石油长庆油田公司第六采油厂,陕西西安710200

出　　处：《石油勘探与开发》2025年第2期488-498,共11页Petroleum Exploration and Development

基　　金：国家自然科学基金面上项目“知识与数据融合的油气储层薄片智能鉴定方法”(42372175);国家自然科学基金卓越研究群体项目“数字经济时代的资源环境管理理论与应用”(72088101);中国石油天然气股份有限公司科技项目“油气勘探开发人工智能关键技术研究”(2023DJ84);中国石油-北京大学战略合作协议基础研究合作项目。

摘　　要：针对现有砂岩岩石结构评价方法依赖于肉眼观察、效率低,且磨圆度等仍处于半定量分析、粒度分析无法分类型统计等问题,提出一种基于视觉大模型(SAM)的砂岩岩石结构智能评价方法。通过设计基于秩-分解矩阵适配器的SAM轻量化微调方法,构建多光谱岩石颗粒分割模型(CoreSAM),实现岩石颗粒边缘提取与类型识别,在此基础上提出一套岩石结构量化评价方法,评价粒度、分选性、磨圆度、颗粒接触关系及胶结类型等指标。实验结果表明,CoreSAM在岩石颗粒分割精度上优于现有方法,且在CT图像、岩心照片等不同类型图像上展现出良好的泛化性,能够实现全样本、分类型的粒度分析以及磨圆度等参数的量化表征,推动储层评价向精准、量化、直观、全面的方向发展。Existing sandstone rock structure evaluation methods rely on visual inspection,with low efficiency,semi-quantitative analysis of roundness,and inability to perform classified statistics in grain size analysis.This study presents an intelligent evaluation method for sandstone rock structure based on the Segment Anything Model(SAM).By developing a lightweight SAM fine-tuning method with rank-decomposition matrix adapters,a multispectral rock particle segmentation model named CoreSAM is constructed,which achieves rock particle edge extraction and type identification.Building upon this,we propose a comprehensive quantitative evaluation system for rock structure,assessing parameters including grain size,sorting,roundness,particle contact and cementation types.The experimental results demonstrate that CoreSAM outperforms existing methods in rock particle segmentation accuracy while showing excellent generalization across different image types such as CT scans and core photographs.The proposed method enables full-sample,classified grain size analysis and quantitative characterization of parameters like roundness,advancing reservoir evaluation towards more precise,quantitative,intuitive,and comprehensive development.

关 键 词：砂岩 岩石结构 智能评价 视觉大模型(SAM) 微调 颗粒边缘提取 类型识别 

分 类 号：TE122.3[石油与天然气工程—油气勘探]