GeoSSL: A geology-aware self-supervised framework for fault detection in 3D seismic data

Abstract

Fault identification is a critical task in 3D seismic interpretation, directly influencing the efficiency and accuracy of reservoir characterization and drilling decisions. However, traditional methods rely heavily on manual experience and high-quality annotated data, making it difficult to adapt to the demands of processing massive amounts of seismic data. To address this, we introduce a novel self-supervised learning (SSL) paradigm specifically designed for geological feature learning, which for the first time unifies masked voxel reconstruction, 3D patch contrastive learning, and multimodal attribute joint contrast into a coherent multi-task pre-training framework. This framework is uniquely tailored to leverage the spatial continuity and physical attributes of seismic data, enabling the model to learn transferable, geologically meaningful representations without any manual labels. It leverages unlabeled seismic data to learn geologically meaningful feature representations and leverages a transfer learning mechanism to achieve high-precision fault identification with small sample sizes. Experiments on multiple public and field-measured datasets, including F3 and SEAM Phase I, demonstrate that this method achieves key metrics such as intersection over union (IoU) and F1-scores of 0.76 and 0.87, respectively, significantly outperforming traditional attribute analysis (IoU = 0.49) and supervised deep learning models (IoU = 0.72). Furthermore, the method remains robust in areas with low signal-to-noise ratios (average confidence > 85%) and consistently estimates fault strike and dip (average absolute error ≤ 3.5°). This research provides an effective solution for reducing reliance on manual annotation and improving the reliability of automated fault interpretation in complex tectonic areas.

Keywords

Self-supervised learning / 3D seismic interpretation / Fault identification / Contrastive learning / Multimodal pre-training

Funding

This project was supported by the Hubei Provincial Engineering Technology Research Center for Information Technology in Clothing Industry Open Fund (Project No.: 2025HBCI03), the Hanjiang Normal University Scientific Research Fund Project (2024B32), and the CCF-Zhipu Large Model Fund Project (202212).

Conflict of interest

The authors declare they have no competing interests.

Author contributions

Conceptualization: Mengge Wang, Junyu Zhang, Haofa Lin Formal analysis: Mengge Wang, Xinrong Hu Investigation: Junyu Zhang, Haofa Lin Methodology: Xinrong Hu, Junyu Zhang Validation: Mengge Wang, Xinrong Hu Writing–original draft: Mengge Wang, Xinrong Hu, Junyu Zhang Writing–review & editing: Mengge Wang, Xinrong Hu, Junyu Zhang, Haofa Lin

Availability of data

The 3D seismic data used in this study were obtained from the publicly available F3 Block dataset, provided by TNO and dGB Earth Sciences via the OpendTect platform. The authors are grateful for the unrestricted data access.

References

[1] Alao JO. The emerging roles of 3D and 4D geophysical and geological modelling in evaluating seismic risks: A critical review. Earthquake Res Adv. 2025; 6(1): 100399. doi: 10.1016/j.eqrea.2025.100399

[2] Seyyedattar M , Zendehboudi S , Butt S. Technical and non—technical challenges of development of offshore petroleum reservoirs: Characterization and production. Nat Resour Res. 2020; 29(3): 2147-2189. doi: 10.1007/s11053—019—09549—7

[3] Daramola GO , Jacks BS , Ajala OA , Akinoso AE. Enhancing oil and gas exploration efficiency through AI—driven seismic imaging and data analysis. Eng Sci Technol J . 2024; 5(4): 1473-1486. doi: 10.51594/estj.v5i4.1077

[4] Verma S , Chopra S , Ha T , Li F. A review of some amplitude—based seismic geometric attributes and their applications. Interpretation. 2022; 10(1): B1-B12. doi: 10.1190/INT—2021—0136.1

[5] Gui J , Chen T , Zhang J , et al. A survey on self—supervised learning: algorithms, applications, and future trends. IEEE Trans Pattern Anal Mach Intell. 2024; 46(12): 9052-9071. doi: 10.1109/TPAMI.2024.3415112

[6] Yang Y , Wang Z , Liu N , et al. Physically driven self—supervised learning and its applications in geophysical inversion. IEEE Trans Geosci Remote Sens. 2024; 62: 4503211. doi: 10.1109/TGRS.2024.3368016

[7] Chai X , Yang T , Gu H , Tang G , Cao W , Wang Y. Geophysics—steered self—supervised learning for deconvolution. Geophys J Int. 2023; 234(1): 40-55. doi: 10.1093/gji/ggad015

[8] Salazar JJ , Maldonado—Cruz E , Ochoa J , Pyrcz MJ. Self—supervised learning for seismic data: Enhancing model interpretability with seismic attributes. IEEE Trans Geosci Remote Sens. 2023; 61: 1-18. doi: 10.1109/TGRS.2023.3285820

[9] Zhang Z , Chen R , Ma J. Improving seismic fault recognition with self—supervised pre—training: a study of 3D transformer—based with multi—scale decoding and fusion. Remote Sens. 2024; 16(5): 922. doi: 10.3390/rs16050922

[10] Sheng H , Wu X , Si X , Li J , Zhang S , Duan X. Seismic foundation model: A next generation deep—learning model in geophysics. Geophysics. 2025; 90(2): IM59-IM79. doi: 10.1190/geo2024—0262.1

[11] Dou Y , Li K. 3D seismic fault detection via contrastive—reconstruction representation learning. Expert Syst Appl. 2024; 249: 123617. doi: 10.1016/j.eswa.2024.123617

[12] Wang J , Ma S , Liu Y , Dong R. AttentionFaultFormer: An attention—enhanced 3D CNN & transformer model for seismic fault detection. J Appl Geophys. 2025; 238: 105707. doi: 10.1016/j.jappgeo.2025.105707

[13] Chen R , Zhang Z , Ma J. Seismic Fault SAM: adapting SAM with lightweight modules and 2.5D strategy for fault detection. In: Yuan B, Ruan Q, Wei S, An G, eds. Proceedings of the 2024 IEEE 17th International Conference on Signal Processing (ICSP2024); October 28—31, 2024; Suzhou, China. IEEE Press; 2024: 436-441. doi: 10.1109/ICSP62129.2024.10846297

[14] Zu S , Zhao P , Ke C , Junxing C . ResACEUnet: an improved transformer Unet model for 3D seismic fault detection. J Geophys Res Mach Learn Comput. 2024; 1(3): e2024JH000232. doi: 10.1029/2024JH000232

[15] Khosro Anjom F , Vaccarino F , Socco LV. Machine learning for seismic exploration: where are we and how far are we from the holy grail? Geophysics. 2024; 89(1): WA157-WA178. doi: 10.1190/geo2023—0129.1

[16] Monteiro BA , Oliveira H , dos Santos JA. Self—supervised learning for seismic image segmentation from few—labeled samples. IEEE Geosci Remote Sens Lett. 2022; 19: 1-5. doi: 10.1109/LGRS.2022.3193567

[17] Choi W , Pyun S , Jou HT. Synthetic training data optimization for enhanced fault detection in seismic images. Lithosphere. 2025; 2025(3): lithosphere_2024_240. doi: 10.2113/2025/lithosphere_2024_240

[18] Noor UA. Machine learning innovations in revolutionizing earthquake engineering: a review. Arch Comput Methods Eng. 2025; 33(1): 687-743. doi: 10.1007/s11831—025—10320—w

[19] Stucchi E , Mazzotti A , Ciuffi S. Seismic preprocessing and amplitude cross—calibration for a time—lapse amplitude study on seismic data from the Oseberg reservoir. Geophys Prospect. 2005; 53(2): 265-282. doi: 10.1111/j.1365—2478.2004.00471.x

[20] Li X , Li K , Xu Z , Huang Z , Dou Y. Fault—Seg—Net: A method for seismic fault segmentation based on multi—scale feature fusion with imbalanced classification. Comput Geotech. 2023; 158: 105412. doi: 10.1016/j.compgeo.2023.105412

[21] Wang W , Huang Q , You S , Yang C , Neumann U. Shape inpainting using 3D generative adversarial network and recurrent convolutional networks. In: Proceedings of the 2017 IEEE International Conference on Computer Vision (ICCV 2017); October 22—29, 2017; Venice, Italy. IEEE; 2017: 2317-2325. doi: 10.1109/ICCV.2017.252

[22] Wang W , Shen S , Yuan Y. Forecasting short—term export volumes with hybrid models integrating SARIMA with attention—based LSTM. Int Sci Tech Econ Res. 2026; 4(1): 1-22. doi: 10.71451/ISTAER2601

[23] Ding Y , Chen G. Joint prediction model of reservoir parameters based on multimodal transformer graph neural operator physical constraint network. Int Sci Tech Econ Res . 2026; 4(1): 70-89. doi: 10.71451/ISTAER2604

[24] Zhao T , Chen G , Pang C , et al. Hybrid convolutional neural network—graph attention network—gradient boosting decision tree model for seismic impedance inversion prediction. J Seismic Explor. 2025; 34(5): 81-98. doi: 10.36922/JSE025310051

[25] Wang Z , Bovik AC , Sheikh HR. Structural similarity based image quality assessment. In: Wu HR, Rao KR, eds. Digital Video Image Quality and Perceptual Coding. CRC Press; 2017: 225-242.

[26] Zheng J , Tang Y , Huang A , Wu D. Hierarchical multivariate representation learning for face sketch recognition. IEEE Trans Emerg Top Comput Intell. 2024; 8(2): 2037-2049. doi: 10.1109/TETCI.2024.3359090

[27] Sun P , Zheng Y , Xu W , Li J , Yang J. Completing missing entities: Exploring consistency reasoning for remote sensing object detection. IEEE Trans Image Process . 2026; 35: 569-584. doi: 10.1109/TIP.2025.3648164

[28] Belmouhcine A , Pham MT , Lefèvre S. YOLO—G3CF: Gaussian contrastive cross—channel fusion for multimodal object detection. IEEE Geosci Remote Sens Lett. 2025; 22: 8002005. doi: 10.1109/LGRS.2025.3564181

[29] Si X , Wu X , Sheng H , Zhu J , Li Z. SeisCLIP: A seismology foundation model pre—trained by multimodal data for multipurpose seismic feature extraction. IEEE Trans Geosci Remote Sens. 2024; 62: 5903713. doi: 10.1109/TGRS.2024.3354456

[30] Contreras V , Stewart JP , Kishida T , et al. NGA—Sub source and path database. Earthq Spectra. 2022; 38(2): 799-840. doi: 10.1177/87552930211065054

[31] Sun X , Xu Q , Yin Z , et al. Prediction equations for pulse parameters based on physics—based ground motion simulation. Appl Geophys. 2025. doi: 10.1007/s11770—025—1311—z

[32] Wu S , Wang Y. Seismic image dip estimation by multiscale principal component analysis. IEEE Trans Geosci Remote Sens. 2022; 61: 5900410. doi: 10.1109/TGRS.2022.3229332

[33] Yang Y , Jin BB , Sun X , et al. Exact counting of subtrees with diameter no more than d in trees: a generating function approach. Inf Comput. 2025; 307: 105353. doi: 10.1016/j.ic.2025.105353

[34] Zhou G , Li T , Li K , Chu S , Zhu X. Fault—GSA: high generalization 3D fault detection method based on sparse annotations. IEEE Trans Geosci Remote Sens. 2025; 63: 5910317. doi: 10.1109/TGRS.2025.3557022