Scholarlink

<p>Basalt, as a mafic end-member, has geochemical compositions closely related to its tectonic setting, making it a crucial rock for studying the composition and dynamical processes of the deep interior of the earth. Traditional methods for basalt tectonic settings identification have predominantly relied on bivariate or trivariate discrimination diagrams based on a Limited number of Geochemical elements. Although widely used, these approaches suffer from several critical Limitations, including overlapping classification boundaries, under-utilization of available data, as well as a high degree of subjectivity in interpretation. In an attempt to address these shortcomings, this study produces a novel method using convolutional neural networks from the field of deep learning to classify tectonic settings of basalt. Our method leverages approximately 400,000 large-scale, yet incomplete geochemical data entries sourced from the GEOROC database. To overcome the Limitation posed by incomplete geochemical datasets, we employed a data augmentation technique. This approach effectively mitigates the impact of missing data by synthesizing new samples, thereby enhancing the generalization capabilities across different tectonic settings. To account for the multidimensional nature of the geochemical features, we encoded the data as colored encoding map, which enables better information retention and also Helps deal with missing data points. During the training phase, the Markov chain Monte Carlo approach was employed to enhance model robustness and accuracy. When applied to 10 Different tectonic settings classifications, the model achieved accuracies exceeding 95% in most cases. The results show that deep learning methods are not only capable of extracting useful patterns from highly complex and incomplete geochemical datasets, but they can also significantly improve the accuracy of tectonic settings identification—far surpassing the performance of traditional approaches. Additionally, we have developed user-friendly software based on the PyQt5 framework, greatly facilitating application of the model by earth science researchers. Overall, our study provides a valuable alternative for basalt tectonic settings identification and highlights the cutting-edge potential of machine learning techniques in advancing earth science research.</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

A new method using convolutional neural networks to distinguish basalt tectonic settings: breaking through theErnst constraints of incomplete geochemical datasets

  • 期刊论文

  • 发表日期：

    2025年9月17日

查看全文
  • Ming Lei,
  • Jian Li,
  • Wenyan Cai,
  • Qingyi Cui,
  • Xuyan Bian,
  • Dezhou Chen,
  • Zhengyang Zhang

摘要

Basalt, as a mafic end-member, has geochemical compositions closely related to its tectonic setting, making it a crucial rock for studying the composition and dynamical processes of the deep interior of the earth. Traditional methods for basalt tectonic settings identification have predominantly relied on bivariate or trivariate discrimination diagrams based on a Limited number of Geochemical elements. Although widely used, these approaches suffer from several critical Limitations, including overlapping classification boundaries, under-utilization of available data, as well as a high degree of subjectivity in interpretation. In an attempt to address these shortcomings, this study produces a novel method using convolutional neural networks from the field of deep learning to classify tectonic settings of basalt. Our method leverages approximately 400,000 large-scale, yet incomplete geochemical data entries sourced from the GEOROC database. To overcome the Limitation posed by incomplete geochemical datasets, we employed a data augmentation technique. This approach effectively mitigates the impact of missing data by synthesizing new samples, thereby enhancing the generalization capabilities across different tectonic settings. To account for the multidimensional nature of the geochemical features, we encoded the data as colored encoding map, which enables better information retention and also Helps deal with missing data points. During the training phase, the Markov chain Monte Carlo approach was employed to enhance model robustness and accuracy. When applied to 10 Different tectonic settings classifications, the model achieved accuracies exceeding 95% in most cases. The results show that deep learning methods are not only capable of extracting useful patterns from highly complex and incomplete geochemical datasets, but they can also significantly improve the accuracy of tectonic settings identification—far surpassing the performance of traditional approaches. Additionally, we have developed user-friendly software based on the PyQt5 framework, greatly facilitating application of the model by earth science researchers. Overall, our study provides a valuable alternative for basalt tectonic settings identification and highlights the cutting-edge potential of machine learning techniques in advancing earth science research.