Identification and characterization of fractured cavities in carbonate reservoirs using Swin-UNet transformer and seismic attribute compression fusion

Abstract

Identifying and characterizing fractured cavities is essential for exploring carbonate reservoirs. However, characterizing the development and distribution of fractured cavities through post-stack seismic attribute analysis remains challenging. Recently, convolutional neural networks (CNNs), such as UNet and its enhanced versions, have enabled the quantitative identification of fractured cavities. Despite these advancements, the local receptive field and weight-sharing mechanisms of these CNNs limit their capability to capture long-range features within strike-slip fault systems. In addition, neural networks are inherently affected by data uncertainty. To address these challenges, a two-step methodology is proposed. The first step utilizes a Swin-UNet transformer (UNETR) model, enhanced with an attention gate, to interpret fractured cavities. The transformer in Swin-UNETR improves the detection of fractured cavities in strike-slip fault zones, whereas the attention gate enhances the recognition of small fractured cavities by increasing their response in the feature maps. This enhanced Swin-UNETR model overcomes the limitations in modeling long-range features. In the second step, the fractured-cavity identification results are combined with seismic attributes from conventional analysis. Principal component analysis is employed both to increase the relative weight of the neural network recognition results in the attribute fusion and to reduce the uncertainty associated with any single identification method. The methodology was validated in the Shunbei area, yielding horizontal segmentation and vertical zonation of fractured cavities, as well as their characterization through fixed-grid modeling. By combining deep learning-based feature extraction with seismic attributes, this approach improves the accuracy of fractured cavity identification and characterization in carbonate reservoirs.

Keywords

Fractured cavity identification and characterization / Seismic attribute compression fusion / Convolutional neural network

Acknowledgments

None.

Funding

This work was supported by the National Natural Science Foundation of China (Grant no. 42274180) and the Graduate Innovation Program of China University of Mining and Technology (Grant no. 2025WLJCRCZL003).

Conflict of interest

Qiang Guo is an Editorial Board Member of this journal, but was not in any way involved in the editorial and peer-review process conducted for this paper, directly or indirectly. The authors declare they have no competing interests.

Author contributions

Conceptualization: Yunhao Cui, Chao Xu

Formal analysis: Zhiqiang Lu

Investigation: Yunhao Cui, Chao Xu, Zhanpeng Chen, Yuwen Qian

Methodology: Yunhao Cui, Yaping Huang, Qiang Guo

Writing-original draft: Yunhao Cui

Writing-review & editing: Yuhua Chen

Availability of data

Data will be made available upon request to the corresponding author.

References

[1] Ma Y, Cai X, Yun L, et al. Practice and theoretical and technical progress in exploration and development of Shunbei ultra-deep carbonate oil and gas field, Tarim Basin, NW China. Pet Explor Dev. 2022; 49(1):1-20. doi: 10.1016/s1876-3804(22)60001-6

[2] Wu X, Yan S, Qi J, Zeng H. Deep learning for characterizing paleokarst collapse features in 3‐D seismic images. JGR Solid Earth. 2020; 125(9):e2020JB019685. doi: 10.1029/2020jb019685

[3] Li Z, Li H, Liu J, Deng G, Gu H, Yan Z. 3D seismic intelligent prediction of fault-controlled fractured-vuggy reservoirs in carbonate reservoirs based on a deep learning method. J Geophys Eng. 2024; 21(2):345-358. doi: 10.1093/jge/gxae005

[4] Gui Z, Zhang J, Zhang Y, Sun C. Characterization of fault-karst reservoirs based on deep learning and attribute fusion. Acta Geophys. 2024; 73(2):1335-1347. doi: 10.1007/s11600-024-01420-5

[5] Zhang G, Lin C, Ren L, et al. Seismic characterization of deeply buried paleocaves based on Bayesian deep learning. J Nat Gas Sci Eng. 2022; 97:104340. doi: 10.1016/j.jngse.2021.104340

[6] Vaswani A, Shazeer N, Parmar N, et al. Attention is All You Need. [arXiv Preprint]; 2023. doi: 10.48550/arXiv.1706.03762

[7] Chen L, Liu X, Zhou H, Zhang H, Lyu F, Mo Q. Identification of carbonate cave reservoirs based on variational bayesian principal component analysis. IEEE Trans Geosci Remote Sens. 2023; 61:1-10. doi: 10.1109/tgrs.2023.3333962

[8] Li B, He Y, Chen W, Shang H, Wang L. Geological modeling of carbonate fracture-cavity reservoir: Case study of Shunbei fault zone No. 5. Front Earth Sci. 2025; 13:1559030. doi: 10.3389/feart.2025.1559030

[9] Wang S, Si X, Cai Z, Sun L, Wang W, Jiang Z. Fast global self-attention for seismic image fault identification. IEEE Trans Geosci Remote Sens. 2024; 62:1-11. doi: 10.1109/tgrs.2024.3436066

[10] Ma Q, Duan T. Multi-level ultra-deep fault-controlled karst reservoirs characterization methods for the Shunbei field. Front Earth Sci. 2023; 11:1149678. doi: 10.3389/feart.2023.1149678

[11] Zheng D, Zhang L, Shen F. Structural heterogeneities in fault-controlled cavity carbonate reservoirs: Insights into fracture and fault patterns. J Struct Geol. 2025; 193:105358. doi: 10.1016/j.jsg.2025.105358

[12] Li H, Lin C, Ren L, Zhang G, Chang L, Dong C. Quantitative prediction of multi-period tectonic fractures based on integrated geological-geophysical and geomechanics data in deep carbonate reservoirs of Halahatang oilfield in northern Tarim Basin. Mar Pet Geol. 2021; 134:105377. doi: 10.1016/j.marpetgeo.2021.105377

[13] Ding Y, Du Q, Fu L, Jian S. Reliability analysis of seismic attribute in the detection of fault-karst. Interpretation. 2020; 8(4):SP61-SP70. doi: 10.1190/int-2019-0299.1

[14] Wang S, Cai Z, Si X, Cui Y. A three-dimensional geological structure modeling framework and its application in machine learning. Math Geosci. 2023; 55(2):163-200. doi: 10.1007/s11004-022-10027-9

[15] Wu X, Liang L, Shi Y, Fomel S. FaultSeg3D: Using synthetic data sets to train an end-to-end convolutional neural network for 3D seismic fault segmentation. Geophys. 2019; 84(3):IM35-IM45. doi: 10.1190/geo2018-0646.1

[16] Liu Z, Lin Y, Cao Y, et al. Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows. [arXiv Preprint]; 2021. p. 10012-10022. doi: 10.48550/arXiv.2103.14030

[17] Hatamizadeh A, Tang Y, Nath V, et al. UNETR: Transformers for 3D Medical Image Segmentation. [arXiv Preprint]; 2021. p. 574-584.doi: 10.48550/arXiv.2103.10504

[18] Oktay O, Schlemper J, Folgoc LL, et al. Attention U-Net: Learning Where to Look for the Pancreas. [arXiv Preprint]; 2018. doi: 10.48550/arXiv.1804.03999

[19] Hatamizadeh A, Nath V, Tang Y, Yang D, Roth H, Xu D. Swin UNETR: Swin Transformers for Semantic Segmentation of Brain Tumors in MRI Images. [arXiv Preprint]; 2022. doi: 10.48550/arXiv.2201.01266

[20] Mishra S, Sarkar U, Taraphder S, et al. Multivariate statistical data analysis- principal component analysis. Int J Livest Res. 2017; 7:60-78. doi: 10.5455/ijlr.20170415115235

[21] Song S, Mukerji T, Hou J, Zhang D, Lyu X. GANSim‐3D for conditional geomodeling: Theory and field application. Water Resour Res. 2022; 58(7):e2021WR031865. doi: 10.1029/2021wr031865