Scholarlink

<p>Geochemical anomalies are crucial for predicting mineral exploration targets. On the one hand, traditional anomaly extraction methods assume that data conform to specific statistical distributions, limiting their ability to accurately capture nonlinear features in complex geochemical data. Data-driven machine learning algorithms (non-deep learning algorithms such as random forest (RF) and support vector machine (SVM), etc.), on the other hand, typically require large amounts of labeled data but often overlook the geological relationships between components, resulting in findings that are difficult to interpret. Unsupervised learning leverages inherent nonlinear dependencies among multivariate geologic variables to elucidate diagnostic signatures through topological analysis in feature space, thereby revealing characteristic assemblages indicative of underlying petrogenetic processes. These methods address the challenges posed by the complex statistical distributions in geochemical data and the over-reliance on known mineral locations in current data-driven approaches. To address these challenges, this research introduces a novel technique, SOMML (self-organizing map and metric learning), which integrates self-organizing map (SOM) with metric learning (ML). The most prominent neurons representing geochemical anomalies are extracted by learning the spatial distribution patterns of mineralization-related indicators using the adaptive weights of SOM. Meanwhile, machine learning metrics leverage the similarity between these anomalous neurons and their surrounding neurons to achieve automatic geochemical anomaly extraction. The findings demonstrate that the SOMML methodology can effectively identify geochemical anomaly locations in the study area and offer significant advantages over traditional approaches, including higher accuracy and geological interpretability. Compared with other data-driven techniques (RF, SVM), the SOMML method identifies the same number of known occurrences within a smaller anomaly region, demonstrating its learning capability and reliability in complex geochemical settings. This method offers a robust solution for the automated extraction of geochemical anomaly information, with broad applicability and important implications for mineral prediction.</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

Automated Geochemical Anomaly Detection Using Self-Organizing Map and Metric Learning

  • 期刊论文

  • 发表日期：

    2025年8月19日

查看全文
  • Yajie Feng,
  • Yongzhi Wang,
  • Cheng Wang,
  • Jiangtao Tian,
  • Zhijie Zeng,
  • Yigao Cheng,
  • ShiBo Wen

摘要

Geochemical anomalies are crucial for predicting mineral exploration targets. On the one hand, traditional anomaly extraction methods assume that data conform to specific statistical distributions, limiting their ability to accurately capture nonlinear features in complex geochemical data. Data-driven machine learning algorithms (non-deep learning algorithms such as random forest (RF) and support vector machine (SVM), etc.), on the other hand, typically require large amounts of labeled data but often overlook the geological relationships between components, resulting in findings that are difficult to interpret. Unsupervised learning leverages inherent nonlinear dependencies among multivariate geologic variables to elucidate diagnostic signatures through topological analysis in feature space, thereby revealing characteristic assemblages indicative of underlying petrogenetic processes. These methods address the challenges posed by the complex statistical distributions in geochemical data and the over-reliance on known mineral locations in current data-driven approaches. To address these challenges, this research introduces a novel technique, SOMML (self-organizing map and metric learning), which integrates self-organizing map (SOM) with metric learning (ML). The most prominent neurons representing geochemical anomalies are extracted by learning the spatial distribution patterns of mineralization-related indicators using the adaptive weights of SOM. Meanwhile, machine learning metrics leverage the similarity between these anomalous neurons and their surrounding neurons to achieve automatic geochemical anomaly extraction. The findings demonstrate that the SOMML methodology can effectively identify geochemical anomaly locations in the study area and offer significant advantages over traditional approaches, including higher accuracy and geological interpretability. Compared with other data-driven techniques (RF, SVM), the SOMML method identifies the same number of known occurrences within a smaller anomaly region, demonstrating its learning capability and reliability in complex geochemical settings. This method offers a robust solution for the automated extraction of geochemical anomaly information, with broad applicability and important implications for mineral prediction.