基于深度学习的碳酸盐岩颗粒分类识别  

Research on classification and recognition of carbonate particles based on deep learning

作　　者：杜崇娇 余晓露[1,2] 彭金宁[1,2] 马中良 王强[1,2] DU Chongjiao;YU Xiaolu;PENG Jinning;MA Zhongliang;WANG Qiang(Key Laboratory of Petroleum Accumulation Mechanisms,SINOPEC,Wuxi 214126,China;Wuxi Research Institute of Petroleum Geology,SINOPEC,Wuxi 214126,China)

机构地区：[1]中国石油化工集团公司油气成藏重点实验室,江苏无锡214126 [2]中国石油化工股份有限公司石油勘探开发研究院无锡石油地质研究所,江苏无锡214126

出　　处：《成都理工大学学报（自然科学版）》2023年第5期629-636,共8页Journal of Chengdu University of Technology: Science & Technology Edition

基　　金：中国石化优秀青年科技创新项目(P19028);中石化股份公司项目(P21042-5)。

摘　　要：针对碳酸盐岩微相分析受人工鉴定经验性和主观性影响,使得传统方法难以准确、客观判识颗粒类型等问题。由此,本文作者提出一种基于深度学习的碳酸盐岩颗粒显微图像识别方法,并以ResNet50为基础网络框架,通过制作数据集、训练模型、预测分类等步骤,设计了一个碳酸盐岩主要颗粒类型自动分类识别模型。利用该模型对生物碎屑、内碎屑、包粒、球粒和团块5种颗粒进行分类识别,再采用混淆矩阵进行评价,结果显示识别准确率达到95%。不仅为碳酸盐岩微相分析提供了新方法,也为深度学习应用于实际碳酸盐岩颗粒分类识别提供了可行性论证,具有一定的实用价值。The grain type of carbonate rocks is not only an important indicator of sedimentary environment and sedimentary cycle,but also a key basis for classification and naming of carbonate rocks.Based on ResNet50 network framework,an automatic classification and recognition model for major particle types of carbonate rock is designed through the steps of making data sets,training models and prediction classification.This model aims to solve the difficulty of accurate and objective identification of particle types by traditional methods and try to avoid the effect of experience and subjectivity of manual identification on carbonate microfacies analysis.Firstly,the model is used to classify and recognize 5 kinds of particles,such as bioclast,intraclast,coated grain,peloid and lump,and then the image recognition effect is evaluated by using confusion matrix.The results show that the accuracy of this model for microscopic image recognition of carbonate particles reaches 95%.Therefore,the model not only provides a new method for carbonate microfacies analysis,but also furnishes a feasibility demonstration for applying deep learning to the classification and identification of actual carbonate particles.

关 键 词：碳酸盐岩颗粒 深度学习 ResNet50 图像识别 

分 类 号：TE344[石油与天然气工程—油气田开发工程] P588[天文地球—岩石学]