基于改进的SKnet和Bi-GRU的岩石薄片图像矿物识别  

Mineral recognition of rock thin section images based on improved SKnet and Bi-GRU

作　　者：刘勇[1] 吴晓红[1] 滕奇志[1] 何海波 LIU Yong;WU Xiaohong;TENG Qizhi;HE Haibo(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China;Chengdu Xitu Technology Co.Ltd,Chengdu 610065,China)

机构地区：[1]四川大学电子信息学院,成都610065 [2]成都西图科技有限公司,成都610065

出　　处：《智能计算机与应用》2023年第1期104-111,共8页Intelligent Computer and Applications

基　　金：国家自然科学基金项目资助(62071315)。

摘　　要：通过分析岩石薄片中矿物成分,研究储集层空间结构,对后续的油气勘探开发具有重要意义。基于正交偏光序列图像的矿物识别研究已经取得了一些成果,但多数方法未利用矿物颗粒在序列图中的变化信息,本文借鉴视频分类的思想,针对岩石矿物颗粒正交偏光序列图像,结合岩石矿物颗粒消光性特点,构建了卷积神经网络和循环神经网络相结合的识别模型。卷积神经网络选用SKnet并在此基础上添加了空间特征融合机制,循环神经网络采用双向门控循环单元(Bidirectional Gated Recurrent Unit,Bi-GRU)来提取序列图像的前后关联特征。选取石英、碱性长石、斜长石、岩屑4类矿物颗粒序列图像构建数据集进行验证,结果表明本文提出的矿物颗粒识别方法识别效果良好。By analyzing the mineral composition of rock slices,we can study the structure of reservoir space,which is of great significance to the subsequent oil and gas exploration and development.Mineral identification research based on orthogonal polarized light sequence images has achieved some results,but most methods do not use the change information of mineral particles in sequence images.In this paper,inspired by video classification,a mineral recognition model combining the convolutional neural network and recurrent neural network is constructed based on the orthogonal polarizing image sequence,which makes use of the extinction characteristics of rock mineral grains.A spatial feature fusion mechanism is added to SKnet as the convolutional neural network and the recurrent neural network is the bidirectional gated recurrent unit(Bi-GRU)to extract the sequence features of mineral grains in the polarization image sequence.We evaluate our method on the dataset containing four types of mineral grains image sequences,including quartz,alkali feldspar,anorthosite,and rock debris.The experimental results demonstrate that the proposed rock mineral recognition method in this paper achieves superior identification performance.

关 键 词：矿物颗粒识别 偏光序列图像 消光性 SKnet 双向门控循环单元 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]