Научная визуализация

Scientific Visualization

Электронный журнал открытого доступа

Национальный Исследовательский Ядерный Университет "МИФИ"

      ISSN 2079-3537      

 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                                                                                             





Научная визуализация, 2026, том 18, номер 3, страницы 41 - 51, DOI: 10.26583/sv.18.3.04

Deep-Learning-Based Workflow for Cavern Segmentation and Derived Profile Construction from Core Photos

Авторы: D.O. Makienko1,A,B, G.A. Kossov2,A, V.V. Abashkin3,A, I.A. Seleznev4,A, I.V. Safonov5,A,B

A LLC STISS, Moscow, Russia

B National Research Nuclear University "MEPhI", Moscow, Russia

1 ORCID: 0000-0001-7341-6128, dmakienko@slb.ru

2 ORCID: 0009-0001-7012-6069, gkossov@slb.ru

3 ORCID: 0009-0003-9827-1780, vabashkin@slb.ru

4 ORCID: 0009-0007-7977-4904, iseleznev@slb.ru

5 ORCID: 0000-0002-8270-943X, isafonov@slb.ru

 

Аннотация

To estimate hydrocarbon reserves and plan production, geologists construct reservoir models that describe lithofacies variations with depth. In carbonate rocks, caverns provide important information about storage capacity and fluid flow potential. Photographs of core samples allow analysis of surface-visible caverns. Segmentation of cavern-like objects in core photographs enables profile construction to identify depth intervals where lithofacies change, reflecting variations in sedimentation conditions and corresponding reservoir properties. Rock heterogeneity, varying imaging conditions, and non-target structures such as fractures and artifacts complicate cavern analysis from photographs. This paper proposes a deep-learning-based workflow for segmentation of cavern-like objects for visu-alization of the cavernosity profile intended to support lithofacies delineation. The work-flow includes semi-automatic annotation preparation for training a deep learning segmen-tation model. Model training relies on augmented images to account for varying bright-ness distributions. The segmentation results enable visualization of the depth-dependent profile that qualitatively reflects the distribution of cavern-like objects and may support description and delineation of lithofacies.

 

Ключевые слова: cavern segmentation, cavernosity visualization, lithofacies delineation.