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

• • 上一篇    下一篇

基于改进的EfficientNet模型的岩屑岩性识别*

董文豪1, 张怀1,2,3†   

  1. 1.地球系统数值模拟与应用全国重点实验室,中国科学院大学地球与行星科学学院,北京 100049;
    2.南方海洋科学与工程广东省实验室(珠海),广东 珠海 519080;
    3.北京燕山地球关键带国家野外科学观测研究站,北京 101408
  • 收稿日期:2025-03-31 修回日期:2025-12-11 发布日期:2025-12-29
  • 通讯作者: †E-mail:hzhang@ucas.ac.cn
  • 基金资助:
    *国家自然科学基金联合基金项目(No.U2239205)和国家重点研发计划(Nos.2020YFA0713400,2020YFA0 713401)资助

Lithology identification of cuttings based on improved EfficientNet model

DONG Wenhao1, ZHANG Huai1,2,3   

  1. 1. State Key Laboratory of Earth System Numerical Modeling and Application, College of Earth and Planetary Sciences, University of Chinese Academy of Sciences, Beijing 100049, China;
    2. Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai), Zhuhai 519080, Guangdong, China;
    3. Beijing Yanshan Earth Critical Zone National Research Station, Beijing 101408, China
  • Received:2025-03-31 Revised:2025-12-11 Published:2025-12-29

摘要: 岩性智能识别技术是优化油气勘探决策的核心环节,其关键在于快速解译录井岩屑图像所蕴含的地质信息。随着录井技术的发展,岩屑图像数量剧增且复杂性增加,急需一种快速且准确的岩性识别方法。本文提出了一种基于改进EfficientNet的深度学习与迁移学习模型,用于8种岩屑的分类识别。与ResNet、Inception等经典神经网络方法相比,EfficientNet的迁移学习模型的泛化能力更强,计算速度更快,识别测试集准确率能达到97.84%。

关键词: 岩性识别, 深度学习, 迁移学习, EfficientNet

Abstract: Intelligent lithology identification technology serves as a critical component in optimizing oil and gas exploration decision-making, with its core focus lying in the rapid interpretation of geological information embedded in mud logging cuttings images. As mud logging technology advances, the volume of obtainable cuttings images has surged dramatically while the complexity and ambiguity of cuttings have intensified, creating an urgent demand for rapid and accurate lithology identification methods in field operations. This study develops an improved EfficientNet-based deep learning and transfer learning model to classify eight common types of cuttings. Compared to classical neural networks like ResNet and Inception, the EfficientNet-based transfer learning model shows better generalization and faster computation, achieving a recognition accuracy of 97.84% on the test dataset.

Key words: lithology identification, deep learning, transfer learning, EfficientNet

中图分类号: