AccScience Publishing / JSE / Online First / DOI: 10.36922/JSE026290142
Cite this article
2
Download
3
Views
Related Info Links
More by Authors Links
Journal Browser
Volume | Year
Issue
Search
News and Announcements
View All
ARTICLE

TSU-former seismic data denoising method based on spatiotemporal attention mechanism

Zhenjing Yao1,2 ,  Lan Qin1,2 ,  Meng Lu3* ,  Jingyi Zhu4 ,  Chenyu Zhang1,2
Show Less
1 School of Emergency Communication and Control Engineering, University of Emergency Management, Sanhe, Hebei , China
2 Hebei Key Laboratory of Seismic Disaster Instrument and Monitoring Technology, Sanhe, Hebei , China
3 Key Laboratory of Drilling Technology in Mine Rescue, Ministry of Emergency Management, Beijing , China
4 International Education College, Beijing University of Chemical Technology, Beijing , China
Received: 17 July 2026 | Revised: 4 October 2026 | Accepted: 16 October 2026 | Published online: 19 November 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/ )
Abstract

Seismic data are frequently contaminated by severe environmental noise, reducing the signal-to-noise ratio (SNR) and degrading subsequent processing and interpretation. Although U-Net-based deep learning has been widely used for denoising, such networks suffer from local receptive fields and an imperfect balance between noise suppression and signal preservation. This paper proposes TSU-Former, a supervised seismic data denoising network built on a symmetric U-Net encoder–decoder with a Transformer bottleneck and a spatiotemporal attention mechanism. An improved convolutional unit with a residual design is integrated into both the encoder and decoder to stabilize training and extract multi-scale local features. The Transformer bottleneck models non-local spatiotemporal dependencies within each input window, preserving the continuity of coherent phase axes. At the network output, a spatiotemporal self-adaptive noise enhancement module recalibrates features along the trace and time dimensions to improve the accuracy of noise prediction. Quantitative evaluations on synthetic and field datasets demonstrate that TSU-Former achieves higher SNR, peak SNR, and structural similarity than five representative denoising baselines. In field-data experiments, the estimated SNR is improved by 13.61 dB relative to the noisy input, and the denoised waveforms retain the principal signal characteristics.

Keywords
Denoising
Deep learning
U-Net
Attention mechanism
Transformer
Funding
None.
Conflict of interest
The authors declare they have no competing interests.
References
  1. Bayram H, Rastgeldi Dogan T, Şahin ÜA, Akdis CA. Environmental and health hazards by massive earthquakes. Allergy. 2023;78(8):2081-2084. doi: 10.1111/all.15736
  2. Mavrouli M, Mavroulis S, Lekkas E, Tsakris A. The impact of earthquakes on public health: a narrative review of infectious diseases in the post-disaster period aiming to disaster risk reduction. Microorganisms. 2023;11(2):419. doi: 10.3390/microorganisms11020419
  3. Jackson J, McKenzie D, Priestley K. Relations between earthquake distributions, geological history, tectonics and rheology on the continents. Philos Trans R Soc A Math Phys Eng Sci. 2021;379(2193):20190412. doi: 10.1098/rsta.2019.0412
  4. Abudeif AM, Abdel Aal GZ, Abdelbaky NF, Abdel Gowad AM, Mohammed MA. Evaluation of engineering site and subsurface structures using seismic refraction tomography: a case study of Abydos site, Sohag governorate, Egypt. Appl Sci. 2023;13(4):2745. doi: 10.3390/app13042745
  5. Mohammed MA, Abudeif AM, Abd El-aal AK. Engineering geotechnical evaluation of soil for foundation purposes using shallow seismic refraction and MASW in 15th Mayo, Egypt. J Afr Earth Sci. 2020;162:103721. doi: 10.1016/j.jafrearsci.2019.103721
  6. Abudeif AM, Raef AE, Abdel Moneim AA, Mohammed MA, Farrag AF. Dynamic geotechnical properties evaluation of a candidate nuclear power plant site (NPP): P- and S-waves seismic refraction technique, North Western Coast, Egypt. Soil Dyn Earthq Eng. 2017;99:124-136. doi: 10.1016/j.soildyn.2017.05.006
  7. Abudeif AM. Seismotectonic analysis of Hamada oil pool, Libya, using Landsat and seismic data. Arab J Geosci. 2015;8(9):6811-6834. doi: 10.1007/s12517-014-1700-8
  8. Dittmann T, Liu Y, Morton Y, Mencin D. Supervised machine learning of high-rate GNSS velocities for earthquake strong motion signals. J Geophys Res Solid Earth. 2022;127(11). doi: 10.1029/2022JB024854
  9. Xiong M, Li Y, Wu N. Random-noise attenuation for seismic data by local parallel radial-trace TFPF. IEEE Trans Geosci Remote Sens. 2014;52(7):4025-4031. doi: 10.1109/TGRS.2013.2278981
  10. Li F, Sun F, Liu N, Xie R. Denoising seismic signal via resampling local applicability functions. IEEE Geosci Remote Sens Lett. 2022;19:1-5. doi: 10.1109/LGRS.2020.3048110
  11. Ding M, Zhou Y, Chi Y. Self-attention generative adversarial network interpolating and denoising seismic signals simultaneously. Remote Sens. 2024;16(2):305. doi: 10.3390/rs16020305
  12. Ma H, Qian Z, Li Y, Lin H, Shao D, Yang B. Noise reduction for desert seismic data using spectral kurtosis adaptive bandpass filter. Acta Geophys. 2019;67:123-131. doi: 10.1007/s11600-018-0232-0
  13. Li ZX, Sun NN, Zeng TS, Zhang H, Tian JQ. Adaptive subtraction based on the matching filter with three frequency bands of modeled multiples for removing seismic multiples. IEEE Geosci Remote Sens Lett. 2023;20:1-5. doi: 10.1109/LGRS.2023.3261059
  14. Liu Q, Jia R, Zhao C, Liu X, Sun H, Zhang X. Face super-resolution reconstruction based on self-attention residual network. IEEE Access. 2020;8:4110-4121. doi: 10.1109/ACCESS.2019.2962790
  15. Lin H, Li Y, Yang B. Recovery of seismic events by time-frequency peak filtering. Proc IEEE Int Conf Image Process. 2007;V:441-V:444. doi: 10.1109/ICIP.2007.4379860
  16. Tian Y, Li Y, Yang B. Variable-eccentricity hyperbolic-trace TFPF for seismic random noise attenuation. IEEE Trans Geosci Remote Sens. 2014;52(10):6449-6458. doi: 10.1109/TGRS.2013.2296603
  17. Wu N, Li Y, Ma H, Xu X. Intermediate-frequency seismic record discrimination by radial trace time-frequency filtering. IEEE Geosci Remote Sens Lett. 2014;11(7):1280-1284. doi: 10.1109/LGRS.2013.2292114
  18. Gaci S. The use of wavelet-based denoising techniques to enhance the first-arrival picking on seismic traces. IEEE Trans Geosci Remote Sens. 2014;52(8):4558-4563. doi: 10.1109/TGRS.2013.2282422
  19. Vargas RN, Paschoarelli Veiga AC. Seismic trace noise reduction by wavelets and double threshold estimation. IET Signal Process. 2017;11(9):1069-1075. doi: 10.1049/iet-spr.2017.0061
  20. Anvari R, Siahsar MAN, Gholtashi S, Kahoo AR, Mohammadi M. Seismic random noise attenuation using synchrosqueezed wavelet transform and low-rank signal matrix approximation. IEEE Trans Geosci Remote Sens. 2017;55(11):6574-6581. doi: 10.1109/TGRS.2017.2730228
  21. Hennenfent G, Fenelon L, Herrmann FJ. Nonequispaced curvelet transform for seismic data reconstruction: a sparsity-promoting approach. Geophysics. 2010;75(6):WB203-WB210. doi: 10.1190/1.3494032
  22. Liu W, Cao S, Li G, He Y. Reconstruction of seismic data with missing traces based on local random sampling and curvelet transform. J Appl Geophys. 2015;115:129-139. doi: 10.1016/j.jappgeo.2015.02.009
  23. Song J, Li Z, Wang G, Lei G, Yang J. Weak seismic signal enhancement using curvelet transform and compressive sampling. IEEE Trans Geosci Remote Sens. 2023;61:5907011. doi: 10.1109/TGRS.2023.3253930
  24. Liu J, Gu Y, Chou Y, Gu J. Seismic random noise reduction using adaptive threshold combined scale and directional characteristics of shearlet transform. IEEE Geosci Remote Sens Lett. 2020;17(9):1637-1641. doi: 10.1109/LGRS.2019.2949806
  25. Gómez JL, Velis DR. A simple method inspired by empirical mode decomposition for denoising seismic data. Geophysics. 2016;81(6):V403-V413. doi: 10.1190/geo2015-0566.1
  26. Han J, van der Baan M. Microseismic and seismic denoising via ensemble empirical mode decomposition and adaptive thresholding. Geophysics. 2015;80(6):KS69-KS80. doi: 10.1190/geo2014-0423.1
  27. Li J, Li Y, Li Y, Qian Z. Downhole microseismic signal denoising via empirical wavelet transform and adaptive thresholding. J Geophys Eng. 2018;15(6):2469-2480. doi: 10.1088/1742-2140/aacf63
  28. Zhang S, Li Y. Seismic exploration desert noise suppression based on complete ensemble empirical mode decomposition with adaptive noise. J Appl Geophys. 2020;180:104055. doi: 10.1016/j.jappgeo.2020.104055
  29. Dragomiretskiy K, Zosso D. Variational mode decomposition. IEEE Trans Signal Process. 2014;62(3):531-544. doi: 10.1109/TSP.2013.2288675
  30. Yu S, Ma J. Complex variational mode decomposition for slope-preserving denoising. IEEE Trans Geosci Remote Sens. 2018;56(1):586-597. doi: 10.1109/TGRS.2017.2751642
  31. Yao X, Zhou Q, Wang C, Hu J, Liu P. An adaptive seismic signal denoising method based on variational mode decomposition. Measurement. 2021;177:109277. doi: 10.1016/j.measurement.2021.109277
  32. Gan S, Chen Y, Zu S, Qu S, Zhong W. Structure-oriented singular value decomposition for random noise attenuation of seismic data. J Geophys Eng. 2015;12(2):262-272. doi: 10.1088/1742-2132/12/2/262
  33. Rekapalli R, Tiwari RK, Sen MK, Vedanti N. 3D seismic data denoising and reconstruction using multichannel time slice singular spectrum analysis. J Appl Geophys. 2017;140:145-153. doi: 10.1016/j.jappgeo.2017.04.001
  34. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436-444. doi: 10.1038/nature14539
  35. Si X, Yuan Y. Random noise attenuation based on residual learning of deep convolutional neural network. SEG Technical Program Expanded Abstracts. 2018:1986-1990. doi: 10.1190/segam2018-2985176.1
  36. Zhao Y, Li Y, Dong X, Yang B. Low-frequency noise suppression method based on improved DnCNN in desert seismic data. IEEE Geosci Remote Sens Lett. 2019;16(5):811-815. doi: 10.1109/LGRS.2018.2882058
  37. Jiang J, Ren H, Zhang M. A convolutional autoencoder method for simultaneous seismic data reconstruction and denoising. IEEE Geosci Remote Sens Lett. 2022;19:7503405. doi: 10.1109/LGRS.2021.3073560
  38. Wang E, Nealon J. Applying machine learning to 3D seismic image denoising and enhancement. Interpretation. 2019;7(3):SE131-SE139. doi: 10.1190/INT-2018-0224.1
  39. Wang R, Zhang R, Bao C, Qiu L, Yang D. Adapting the residual dense network for seismic data denoising and upscaling. Geophysics. 2022;87(4):V321-V340. doi: 10.1190/geo2021-0294.1
  40. Li Y, Wang Y, Wu N. Noise suppression method based on multi-scale dilated convolution network in desert seismic data. Comput Geosci. 2021;156:104910. doi: 10.1016/j.cageo.2021.104910
  41. Gao Z, Zhang S, Cai J, Hong L, Zheng J. Research on deep convolutional neural network time-frequency domain seismic signal denoising combined with residual dense blocks. Front Earth Sci. 2021;9:681869. doi: 10.3389/feart.2021.681869
  42. Feng J, Li X, Liu X, Chen C. Multigranularity feature fusion convolutional neural network for seismic data denoising. IEEE Trans Geosci Remote Sens. 2022;60:5907911. doi: 10.1109/TGRS.2021.3123509
  43. Liu Y, Li CF, Li G. Active-source ocean bottom seismometer data denoising based on multiresolution convolutional neural networks. Proc SEG Workshop, 4th Int Workshop Math Geophys. 2022:17-20. doi: 10.1190/iwmg2021-05.1
  44. Tan M, Le QV. MixConv: mixed depthwise convolutional kernels. arXiv. arXiv:1907.09595. doi: 10.48550/arXiv.1907.09595
  45. Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16x16 words: transformers for image recognition at scale. arXiv. arXiv:2010.11929. doi: 10.48550/arXiv.2010.11929
  46. Li F, Liu H, Wang W, Ma J. Swin transformer for seismic denoising. IEEE Geosci Remote Sens Lett. 2024;21:1-5. doi: 10.1109/LGRS.2024.3358234
  47. Chen J, Chen G, Li J, et al. Efficient seismic data denoising via deep learning with improved MCA-SCUNet. IEEE Trans Geosci Remote Sens. 2024;62:1-14. doi: 10.1109/TGRS.2024.3355972
  48. Qi Y, Chen G, Chen J, et al. Seismic denoising via multiround SCU-Net. IEEE Geosci Remote Sens Lett. 2026;23:1-5. doi: 10.1109/LGRS.2025.3640968
  49. Saad, Ravasi, Alkhalifah. Self-supervised multi-stage deep learning network for seismic data denoising. Artif Intell Geosci. 2025;6(1):100123. doi: 10.1016/j.aiig.2025.100123
  50. Saad, Alkhalifah. Distributed acoustic sensing denoising using a self-supervised conditional diffusion model. Geophys Prospect. 2026;74(3). doi: 10.1111/1365-2478.70154
  51. Saad OM, Chen Y, Alkhalifah T. U-Trans: a foundation model for seismic waveform representation and enhanced downstream earthquake tasks. Sci Rep. 2026. doi: 10.1038/s41598-026-41454-x
  52. Xu Z, Luo Y, Wu B, Meng D. S2S-WTV: seismic data noise attenuation using weighted total variation regularized self-supervised learning. IEEE Trans Geosci Remote Sens. 2023;61:1-15. doi: 10.1109/TGRS.2023.3268554
  53. Chen Y, Savvaidis A, Saad OM, et al. TXED: the Texas earthquake dataset for AI. Seismol Res Lett. 2024;95(3):2013-2022. doi: 10.1785/0220230327
Share
Back to top
Journal of Seismic Exploration, Print ISSN: 0963-0651, Published by AccScience Publishing