Scholarlink

<p>The scarcity of training samples is one of the critical challenges that need to be addressed for current automatic labeling methods. Existing methods for lithofacies labeling rely heavily on manual labeling, leading to high costs and low efficiency. Given this, a Bidirectional Long and Short-term Memory Networks and Gaussian Mixture Model method was proposed in this paper, aiming to overcome the low identification accuracy in the small sample scenario of lithofacies. Firstly, logging data was processed using standardization and normalization techniques to optimize data quality. Then, Gaussian Mixture Model clustering was utilized to generate labels, providing supervised labels for model iteration. Finally, a Bidirectional Long and Short-term Memory model was constructed to extract features from logging data and output lithofacies types. Experimental results show that the model achieves an overall identification accuracy of 89%. In sample labeling experiments, the labeling time was significantly reduced, saving about 4.5&#xa0;hours compared to manual labeling. This method achieves efficient and accurate automatic labeling while advancing the standardization of intelligent facies identification for tight sandstone. More importantly, it opens new pathways for the automation and intelligence development in the field of geological exploration, while providing innovative ideas for the interdisciplinary integration of computer science and geology.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An automatic Lithofacies sample labeling method by integrating deep learning and Gaussian mixture model

  • 期刊论文

  • 发表日期：

    2025年6月4日

查看全文
  • Zihao Mu,
  • Chunsheng Li,
  • Zongbao Liu,
  • Tao Liu,
  • Kejia Zhang,
  • Haoran Wang,
  • Haiwei Mu,
  • Yuchen Yang

摘要

The scarcity of training samples is one of the critical challenges that need to be addressed for current automatic labeling methods. Existing methods for lithofacies labeling rely heavily on manual labeling, leading to high costs and low efficiency. Given this, a Bidirectional Long and Short-term Memory Networks and Gaussian Mixture Model method was proposed in this paper, aiming to overcome the low identification accuracy in the small sample scenario of lithofacies. Firstly, logging data was processed using standardization and normalization techniques to optimize data quality. Then, Gaussian Mixture Model clustering was utilized to generate labels, providing supervised labels for model iteration. Finally, a Bidirectional Long and Short-term Memory model was constructed to extract features from logging data and output lithofacies types. Experimental results show that the model achieves an overall identification accuracy of 89%. In sample labeling experiments, the labeling time was significantly reduced, saving about 4.5 hours compared to manual labeling. This method achieves efficient and accurate automatic labeling while advancing the standardization of intelligent facies identification for tight sandstone. More importantly, it opens new pathways for the automation and intelligence development in the field of geological exploration, while providing innovative ideas for the interdisciplinary integration of computer science and geology.