AccScience Publishing / JSE / Online First / DOI: 10.36922/JSE026120051
Submit to JSE
Cite this article
6
Download
350
Views
Journal Browser
Volume | Year
Issue
Search
Forthcoming Issue
Current Issue
View All
News and Announcements
View All
ARTICLE

Research on an automatic seismic first-arrival picking application based on deep learning semantic segmentation

Daicheng Peng1,2,3* ,  Bo Deng2 ,  Guilong Gao2 ,  Xiurong Li2 ,  Gang Wang2 ,  Min Yin2 ,  Xiong Ma1,2*
Show Less
1 Key Laboratory of Exploration Technologies for Oil and Gas Resource, Ministry of Education, Yangtze University, Wuhan, Hubei, China
2 SINOPEC Geophysical Corp. Jianghan Branch, Qianjiang, Hubei, China
3 Hubei Subsurface Multi-scale Imaging Key Laboratory, School of Geophysics and Geomatics, China University of Geosciences, Wuhan, Hubei, China
JSE, 026120051 https://doi.org/10.36922/JSE026120051
Received: 20 March 2026 | Revised: 17 April 2026 | Accepted: 22 April 2026 | Published online: 21 May 2026
© 2026 by the Author(s). This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution 4.0 International License ( https://creativecommons.org/licenses/by/4.0/ )
Download PDF
XML
Cite
Abstract

First-arrival picking is a fundamental step in seismic data processing, directly influencing the accuracy of subsequent procedures such as static correction, near-surface velocity modeling, and seismic imaging. However, under complex near-surface conditions and low signal-to-noise ratio environments, conventional automatic picking methods often suffer from limited accuracy, poor continuity, and strong dependence on data quality, necessitating extensive manual intervention. To address these challenges, this study proposes a semantic segmentation-based deep learning framework for automatic first-arrival picking, reformulating the task as a pixel-level classification problem on two-dimensional time–trace seismic records, enabling holistic modeling of wavefield characteristics before and after the first-arrival. Within this framework, three representative architectures—U-Net, MobileNet-U-Net, and Attention-U-Net—were constructed and systematically evaluated from the perspectives of standard design, lightweight optimization, and attention enhancement. Synthetic seismic datasets generated via finite-difference forward modeling were used for training, and data augmentation strategies were introduced to improve robustness under noise contamination and missing-trace conditions. The trained models were validated on both synthetic test data and real seismic data from the Loess Plateau. The results demonstrate that all three models effectively performed automatic first-arrival picking, with the Attention-U-Net achieving the best overall performance, exhibiting superior accuracy, continuity, and robustness, particularly under strong noise and severe trace-loss conditions. Compared with conventional methods, the proposed approach produced more reliable and stable picking results and significantly reduced the need for manual correction. These findings indicate that semantic segmentation-based deep learning methods, especially those incorporating attention mechanisms, provide an effective and practical solution for high-precision first-arrival picking in complex seismic environments.

Keywords
First-arrival picking
Semantic segmentation
Deep learning
Attention-U-Net
Seismic data processing
Loess Plateau
Funding
This work was supported by the Sinopec Petroleum Engineering Geophysics Co., Ltd. Wuhan Exploration Branch 2025 Research Project (No. CIBG250096); the Open Funds for Hubei Subsurface Multi-scale Imaging Key Laboratory, School of Geophysics and Geomatics, China University of Geosciences (No. SMIL-2025-01); the Youth Project of Hubei Provincial Natural Science Foundation (No. 2025AFB409); the China Postdoctoral Science Foundation (No. 2025M780437); and the Open Fund of the Key Laboratory of Exploration Technologies for Oil and Gas Resources (Yangtze University), Ministry of Education, (No. K2023-04).
Conflict of interest
All authors are employees of SINOPEC Geophysical Corp. Jianghan Branch; however, they were not involved in any activities that could constitute a conflict of interest in relation to this study.
References
  1. Yuan P, Wang S, Hu W, et al. A robust first-arrival picking workflow using convolutional and recurrent neural networks. Geophysics. 2020;85(5):U109–U119. doi: 10.1190/geo2019-0437.1

 

  1. Tsai K C, Hu W, Wu X, et al. Automatic first-arrival picking via deep learning with human interactive learning. IEEE Trans Geosci Remote Sens. 2020;58(2):1380–1391. doi: 10.1109/TGRS.2019.2946118

 

  1. Sabbione J I, Velis D. Automatic first-breaks picking: New strategies and algorithms. Geophysics. 2010;75(4):V67–V76. doi: 10.1190/1.3463703

 

  1. Kahrizi A, Hashemi H. Neuron curve as a tool for performance evaluation of MLP and RBF architecture in first break picking of seismic data. J Appl Geophys. 2014;108:159– 166. doi: 10.1016/j.jappgeo.2014.06.012

 

  1. Allen RV. Automatic earthquake recognition and timing from single traces. Bull Seismol Soc Am. 1978;68(5):1521– 1532. doi: 10.1785/BSSA0680051521

 

  1. Cheng F, Peng D, Qu N, et al. Research on the method of first arrival time picking of microseismic signals and inverse location based on underground water-sealed oil reservoir. Rock Mech Rock Eng. 2025;58(12):13699–13722. doi: 10.1007/s00603-025-04812-1

 

  1. Saragiotis CD, Hadjileontiadis LJ, Panas SM. PAI-S/K: A robust automatic seismic P-phase arrival identification scheme. IEEE Trans Geosci Remote Sens. 2002;40(6):1395– 1404. doi: 10.1109/TGRS.2002.800438

 

  1. Chen Z, Liu H, Zhao J, et al. Real-time event detection based on STA/LTA method using field synchrophasor measurements. IEEE Trans Power Deliv. 2023;38(6):4070– 4080. doi: 10.1109/TPWRD.2023.3301130

 

  1. Peng D, Yang D, Cheng F, et al. 3-D elastic wave simulation and microseismic source localization in underground water-sealed oil-storage caverns. IEEE Trans Geosci Remote Sens. 2026;64:1–18. doi: 10.1109/TGRS.2026.3653856

 

  1. Zhang H, Thurber C, Rowe C. Automatic P-wave arrival detection and picking with multiscale wavelet analysis for single-component recordings. Bull Seismol Soc Am. 2003;93(5):1904–1912. doi: 10.1785/0120020241

 

  1. Akram J, Eaton DW. A review and appraisal of arrival-time picking methods for downhole microseismic data. Geophysics. 2016;81(2):KS71–KS91. doi: 10.1190/geo2014-0500.1

 

  1. Sharma BK, Kumar A, Murthy VM. Evaluation of seismic events detection algorithms. J Geol Soc India. 2010;75(3):533–538. doi: 10.1007/s12594-010-0042-8

 

  1. Dai H, MacBeth C. The application of back-propagation neural network to automatic picking seismic arrivals from single-component recordings. J Geophys Res Solid Earth. 1997;102(B7):15105–15113. doi: 10.1029/97JB00625

 

  1. Gentili S, Michelini A. Automatic picking of P and S phases using a neural tree. J Seismol. 2006;10(1):39–63. doi: 10.1007/s10950-006-2296-6

 

  1. Ross ZE, Meier MA, Hauksson E. P-wave arrival picking and first-motion polarity determination with deep learning. J Geophys Res Solid Earth. 2018;123(6):5120–5129. doi: 10.1029/2017JB015251

 

  1. Guo C, Zhu T, Gao Y, et al. AEnet: Automatic picking of P-wave first arrivals using deep learning. IEEE Trans Geosci Remote Sens. 2021;59(6):5293–5303. doi: 10.1109/TGRS.2020.3010541

 

  1. Krizhevsky A, Sutskever I, Hinton GE. ImageNet classification with deep convolutional neural networks. Commun ACM. 2017;60(6):84-90. doi: 10.1145/3065386

 

  1. Parcollet T, Zhang Y, Morchid M, et al. Quaternion convolutional neural networks for end-to-end automatic speech recognition. arXiv. Preprint posted online 2018. doi: 10.48550/arXiv.1806.07789

 

  1. Socher R, Huang E, Pennin J, et al. Dynamic pooling and unfolding recursive autoencoders for paraphrase detection. In: Advances in Neural Information Processing Systems. NeurIPS Proceedings;2011;24.

 

  1. Hu G, Hu Z, Liu J, et al. Seismic fault interpretation using deep learning-based semantic segmentation method. IEEE Geosci Remote Sens Lett. 2022;19:1–5. doi: 10.1109/LGRS.2020.3041301

 

  1. Wang Y, Qiu Q, Lan Z, et al. Identifying microseismic events using a dual-channel CNN with wavelet packets decomposition coefficients. Comput Geosci. 2022;166:105164. doi: 10.1016/j.cageo.2022.105164

 

  1. Zhang Z, Yan Z, Li H, et al. 3D seismic fault detection using fault orthogonal annotation and barely supervised learning. IEEE Trans Geosci Remote Sens. 2025;63:1–11. doi: 10.1109/TGRS.2025.3531493

 

  1. 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 Seism Explor. 2025;34(5):81–98. doi: 10.36922/JSE025310051

 

  1. Sun P, Zheng J, Zhao J, et al. Fully automatic differentiation with coupling deep neural networks for full-waveform inversion. J Seism Explor. 2025;34(6):16–28. doi: 10.36922/JSE025410085

 

  1. Yuan S, Liu J, Wang S, et al. Seismic waveform classification and first-break picking using convolution neural networks. IEEE Geosci Remote Sens Lett. 2018;15(2):272–276. doi: 10.1109/LGRS.2017.2785834

 

  1. Qin Z, Zheng X, Wang W. A precise picking method for seismic first arrivals based on the residual long short-term memory network driven by time-frequency dual-domain data. J Seism Explor. 2025;34(6):1–15. doi: 10.36922/JSE025330058

 

  1. Zheng J, Lu J, Peng S, et al. An automatic microseismic or acoustic emission arrival identification scheme with deep recurrent neural networks. Geophys J Int. 2018;212(2):1389– 1397. doi: 10.1093/gji/ggx487

 

  1. Wang J, Xiao Z, Liu C, et al. Deep learning for picking seismic arrival times. J Geophys Res Solid Earth. 2019;124(7):6612– 6624. doi: 10.1029/2019JB017536

 

  1. Saad OM, Chen Y, Savvaidis A, et al. Unsupervised deep learning for single-channel earthquake data denoising and its applications in event detection and fully automatic location. IEEE Trans Geosci Remote Sens. 2022;60:1–10. doi: 10.1109/TGRS.2022.3209932

 

  1. Zhu W, Beroza GC. PhaseNet: A deep-neural-network-based seismic arrival-time picking method. Geophys J Int. 2019;216(1):261–273. doi: 10.1093/gji/ggy423

 

  1. Wu Y, Lin Y, Zhou Z, et al. DeepDetect: A cascaded region-based densely connected network for seismic event detection. IEEE Trans Geosci Remote Sens. 2019;57(1):62– 75. doi: 10.1109/TGRS.2018.2852302

 

  1. Ross ZE, Meier MA, Hauksson E, et al. Generalized seismic phase detection with deep learning. Bull Seismol Soc Am. 2018;108(5A):2894–2901. doi: 10.1785/0120180080

 

  1. Wu H, Zhang B, Li F, et al. Semiautomatic first-arrival picking of microseismic events by using the pixel-wise convolutional image segmentation method. Geophysics. 2019;84(3):V143–V155. doi: 10.1190/geo2018-0389.1

 

  1. Badrinarayanan V, Kendall A, Cipolla R. SegNet: A deep convolutional encoder–decoder architecture for image segmentation. IEEE Trans Pattern Anal Mach Intell. 2017;39(12):2481–2495. doi: 10.1109/TPAMI.2016.2644615

 

  1. Wang H, Zhang J, Zhao Z, et al. Automatic velocity picking using a multi-information fusion deep semantic segmentation network. IEEE Trans Geosci Remote Sens. 2022;60:1–10. doi: 10.1109/TGRS.2022.3188669

 

  1. Cui Y, Chen Y, Xu C, et al. Identification and characterization of fractured cavities in carbonate reservoirs using Swin- UNet transformer and seismic attribute compression fusion. J Seism Explor. 2026;35(1):119–138. doi: 10.36922/JSE025420090

 

  1. Wu M, Wu J, Bai M, et al. U-STDRNet: A unified model integrating Swin Transformer and residual dense network for seismic image super-resolution and denoising. J Seism Explor. 2026;35(1):14–32. doi: 10.36922/JSE025400081

 

  1. Wang K, Ma K, Tang C, et al. Study on automatic first-arrival picking of mine microseismic waveforms based on the improved U-Net. Tunn Undergr Space Technol. 2025;166:106995. doi: 10.1016/j.tust.2025.106995

 

  1. Hu L, Zheng X, Duan Y, et al. First-arrival picking with a U-Net convolutional network. Geophysics. 2019;84(6):U45– U57. doi: 10.1190/geo2018-0688.1

 

  1. Han S, Liu Y, Li Y, et al. First-arrival traveltime picking through 3-D U-Net. IEEE Geosci Remote Sens Lett. 2022;19:1–5. doi: 10.1109/LGRS.2021.3096572

 

  1. Chen G, Zhao T, Pang C, et al. An attention-guided graph neural network and U-Net++-based reservoir porosity prediction system. J Seism Explor. 2025;34(4):70–87. doi: 10.36922/JSE025300044

 

  1. Hu G, Li Y, Yang S, et al. Intelligent suppression of marine seismic multiples using deep learning methods. J Ocean Univ China. 2025;24(4):967–978. doi: 10.1007/s11802-025-6016-7

 

  1. Wang H, Zhang J, Wei X, et al. UPNet: Uncertainty-based picking deep learning network for robust first-break picking. IEEE Trans Geosci Remote Sens. 2024;62:1–14. doi: 10.1109/TGRS.2024.3439685

 

  1. Wang H, Zhang J, Wei X, et al. MSSPN: Automatic first-arrival picking using a multistage segmentation picking network. Geophysics. 2024;89(3):U53–U70. doi: 10.1190/geo2023-0110.1

 

  1. Shen T, Jiang X, Wang S, et al. Improved U-Net3+ network for first-arrival picking of noisy earthquake recordings. IEEE Trans Geosci Remote Sens. 2024;62:1–11. doi: 10.1109/TGRS.2024.3387418

 

  1. Wu X, Liang L, Shi Y, et al. FaultSeg3D: Using synthetic data sets to train an end-to-end convolutional neural network for 3D seismic fault segmentation. Geophysics. 2019;84(3):IM35–IM45. doi: 10.1190/geo2018-0646.1
Share
Back to top
Journal of Seismic Exploration, Electronic ISSN: 0963-0651 Print ISSN: 0963-0651, Published by AccScience Publishing
We use cookies on our website to ensure you get the best experience.
Read more about our cookies here
Accept