EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multi-task learning for structural boundary estimation from gravity data

Xiang Huang, Lanqing Chen, Huaisheng Xu, Zhuo Jia

Machine Learning: Science and Technology · 01.08.2026 · 7 (4) · 045037

DOI: 10.1088/2632-2153/ae85ba ↗

Kurz gefasst

Fachartikel in der Zeitschrift Machine Learning: Science and Technology, online erschienen am 28.07.2026. Verfasst von Xiang Huang und 3 weiteren Autor:innen. Die Arbeit nennt 59 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Hunan Provincial Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Euler deconvolution remains one of the most widely used techniques for gravity interpretation because of its simplicity and computational efficiency. However, its outputs are generally limited to discrete source-point estimations and may become unstable under noisy conditions or inappropriate structural-index assumptions, particularly in complex geological environments. To overcome these limitations, we propose a multi-task deep learning framework that directly estimates three key structural depth attributes—the top boundary, structural center, and an effective lower depth—from gravity data. Unlike conventional workflows that first invert for physical properties and subsequently infer structural boundaries, the proposed framework adopts a geometry-driven learning paradigm that directly predicts structurally meaningful depth attributes from gravity anomalies. The shared encoder and multi-decoder architecture enables joint learning of correlated depth attributes, improving depth stability and structural consistency. Synthetic experiments demonstrate that the proposed framework produces spatially continuous and noise-robust structural representations with improved structural coherence and geological interpretability. The results suggest that deep learning provides a geometry-oriented alternative for extracting continuous structural representations from gravity data, rather than serving as a strict mathematical deconvolution method. This work provides a new pathway for potential-field interpretation by offering a direct and stable framework for subsurface structural characterization from gravity anomalies.

Förderung

  • National Natural Science Foundation of China
  • Hunan Provincial Natural Science Foundation of China

Zitieren

APA

Huang, X., Chen, L., Xu, H., & Jia, Z. (2026). Multi-task learning for structural boundary estimation from gravity data. Machine Learning: Science and Technology, 7(4), 045037. https://doi.org/10.1088/2632-2153/ae85ba

Vancouver

Huang X, Chen L, Xu H, Jia Z. Multi-task learning for structural boundary estimation from gravity data. Machine Learning: Science and Technology. 2026;7(4):045037. doi:10.1088/2632-2153/ae85ba

BibTeX
@article{Huang2026,
  author = {Huang, Xiang and Chen, Lanqing and Xu, Huaisheng and Jia, Zhuo},
  title = {Multi-task learning for structural boundary estimation from gravity data},
  journal = {Machine Learning: Science and Technology},
  year = {2026},
  volume = {7},
  number = {4},
  pages = {045037},
  doi = {10.1088/2632-2153/ae85ba}
}
Kostenlos starten →