Scholarlink

<p>Three-dimensional (3D) mineral prospectivity modeling (MPM) focuses on deeper and peripheral targets of known deposits, while complex mineralization processes can lead to variations in 3D digital features of key ore-controlling factors in deeper and peripheral areas compared to their shallower counterparts. The rarity of mineralization events results in an imbalanced distribution of positive and negative samples for 3D MPM. Global domain adaptation may lead to misclassification when addressing the aforementioned imbalanced subdomain shift. To address this issue, this paper proposes a fine-grained multi-kernel maximum mean discrepancy loss (MK-MMD) and constructs a deep subdomain adaptive network (DSAN) based on it. In the 3D MPM case study of the Damiao Fe–V–Ti belt, the performance of the random forest, 3D LeNet5, domain adaptation network, and DSAN models was evaluated. The results demonstrated the superiority of the DSAN model, highlighting that the deep subsurface of the Damiao anorthosite complex holds high mineralization potential.</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

Deep Subdomain Adaptation Network for Three-Dimensional Mineral Prospectivity Modeling with Imbalanced Data: A Case Study of the Damiao–Hongshila Fe–V–Ti Belt, China

  • 期刊论文

  • 发表日期：

    2025年8月17日

查看全文
  • Zhiqiang Zhang,
  • Wenliang Chen,
  • Emmanuel John M. Carranza,
  • Wei Han,
  • Xinxing Liu,
  • Yingjie Li,
  • Juan Zhang,
  • Feng Li,
  • Ziyang Lu,
  • Yongjun Su,
  • Gongwen Wang

摘要

Three-dimensional (3D) mineral prospectivity modeling (MPM) focuses on deeper and peripheral targets of known deposits, while complex mineralization processes can lead to variations in 3D digital features of key ore-controlling factors in deeper and peripheral areas compared to their shallower counterparts. The rarity of mineralization events results in an imbalanced distribution of positive and negative samples for 3D MPM. Global domain adaptation may lead to misclassification when addressing the aforementioned imbalanced subdomain shift. To address this issue, this paper proposes a fine-grained multi-kernel maximum mean discrepancy loss (MK-MMD) and constructs a deep subdomain adaptive network (DSAN) based on it. In the 3D MPM case study of the Damiao Fe–V–Ti belt, the performance of the random forest, 3D LeNet5, domain adaptation network, and DSAN models was evaluated. The results demonstrated the superiority of the DSAN model, highlighting that the deep subsurface of the Damiao anorthosite complex holds high mineralization potential.