Scholarlink

<p>The intrinsic opacity of artificial intelligence (AI) frameworks for mineral prospectivity mapping remains a limitation of their geological applicability and interpretability. To address this, we introduce a spatial-aware parametric UMAP network (SAPUNet), an end-to-end deep learning framework designed to bridge this interpretability gap. SAPUNet integrates a parametric UMAP encoder and a classifier, jointly optimized via a hybrid loss function that synergistically enhances classification accuracy, data topology preservation, and spatial coherence grounded in geological first principles. Applied to the complex Duolong mineral district, the SAPUNet successfully deconstructs superimposed mineralization patterns and generates a geologically meaningful latent space where dimensions correspond directly to ore-forming processes. This framework provides a new paradigm for building trustworthy AI models that augment, rather than replace, geological reasoning.</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

Spatial-Aware Parametric UMAP Network: An Interpretable Deep Learning Analytical Framework for Mineral Prospectivity Mapping

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Yixiao Wu,
  • Wenlei Wang,
  • Changjiang Yuan

摘要

The intrinsic opacity of artificial intelligence (AI) frameworks for mineral prospectivity mapping remains a limitation of their geological applicability and interpretability. To address this, we introduce a spatial-aware parametric UMAP network (SAPUNet), an end-to-end deep learning framework designed to bridge this interpretability gap. SAPUNet integrates a parametric UMAP encoder and a classifier, jointly optimized via a hybrid loss function that synergistically enhances classification accuracy, data topology preservation, and spatial coherence grounded in geological first principles. Applied to the complex Duolong mineral district, the SAPUNet successfully deconstructs superimposed mineralization patterns and generates a geologically meaningful latent space where dimensions correspond directly to ore-forming processes. This framework provides a new paradigm for building trustworthy AI models that augment, rather than replace, geological reasoning.