欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2023, Vol. 40 ›› Issue (6): 743-750.DOI: 10.7523/j.ucas.2022.026

• 地质与地球科学 • 上一篇    下一篇

基于迁移学习的岩屑岩性识别

董文豪, 张怀   

  1. 中国科学院大学地球与行星科学学院 中国科学院计算地球动力学重点实验室, 北京 100049
  • 收稿日期:2022-01-21 修回日期:2022-03-31 发布日期:2022-04-07
  • 通讯作者: 张怀,E-mail:hzhang@ucas.ac.cn
  • 基金资助:
    国家重点研发计划重点专项(2020YFA0713400)、国家杰出青年科学基金(41725017)和国家重大科技基础设施项目资助

Lithology recognition of cuttings based on transfer learning

DONG Wenhao, ZHANG Huai   

  1. CAS Key Laboratory of Computational Geodynamics, College of Earth and Planetary Sciences, University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2022-01-21 Revised:2022-03-31 Published:2022-04-07

摘要: 岩屑录井在地质构造研究及油气勘探等领域中有重要作用。随着录井技术提升,岩屑录井图片数量剧增,传统人工识别岩屑已远不能满足实际工作需求。基于迁移学习的卷积神经网络在图片分类识别中以高效著称。以常见的18种岩屑为研究对象,基于在ImageNet图像数据集上训练好的VGG-16模型建立符合岩屑图片数据集特征的迁移学习模型,并应用到实际的岩性识别中。选用5 877张岩屑录井图片,以3∶1∶1的比例随机划分训练集、验证集和测试集,其岩性识别准确率分别达到99.7%、87.2%和87.3%。测试学习结果表明该方法比卷积神经网络模型在岩性分类识别中具有更高的准确率。

关键词: 岩性识别, 卷积神经网络, 迁移学习, VGG-16

Abstract: Cutting logging plays an important role in the fields of geological structure research and oil and gas exploration. With the improvement of logging technology, the number of cutting logging pictures has increased sharply, and traditional manual identification of cuttings is far from meeting actual work requirements. Convolutional neural networks based on transfer learning are known for their high efficiency in image classification and recognition. This paper focuses on 18 common kinds of cuttings as the research object. Based on the VGG-16 model trained on the ImageNet image data set, a migration learning model conforming to the characteristics of the cuttings image data set is established and applied to the actual lithology recognition. This paper selects 5 877 cutting logging pictures, and the training set, validation set, and test set were randomly divided in the ratio of 3:1:1. The lithology recognition accuracies of the training set, validation set, and test set reach 98.6%, 87.2%, and 87.2%, respectively. The test results on learning show that this method is very effective in lithology classification and recognition.

Key words: lithology recognition, convolution neural network, transfer learning, VGG-16

中图分类号: