TFDenoiser-Edge: A hybrid convolutional neural network–Transformer framework for real-time seismic denoising on edge devices in extreme environments

Abstract

Seismic monitoring in extreme environments, such as arid regions adjacent to the Alxa Desert, faces significant challenges due to complex noise interference from dust storms, high wind noise, and thermal variations. This paper presents TFDenoiser-Edge, a novel hybrid deep learning framework that combines convolutional neural networks (CNN) and Transformer architectures for real-time seismic signal denoising on resource-constrained edge devices. The proposed model employs a U-Net encoder–decoder structure with Transformer modules for global feature modeling in the time-frequency domain. To enable deployment on edge neural processing units (NPUs) with limited memory (≤512 MB), we introduced a mixed-precision quantization strategy that applies INT8 quantization to CNN layers while maintaining BF16 precision for Transformer layers, achieving 3.6× model compression with only 0.3 dB signal-to-noise ratio (SNR) loss. Additionally, a block-wise computation approach reduces peak memory consumption from 86 MB to 7.8 MB. Extensive experiments on Gansu seismic data demonstrated that TFDenoiser-Edge achieved an average SNR improvement of 8.5 dB, with P-wave and S-wave detection rates increasing from 65% to 91% and 52% to 85%, respectively. The model achieved real-time inference with 68 ms latency on edge NPUs while consuming less than 5 W of power, making it suitable for autonomous seismic monitoring in arid and desert regions. The proposed framework demonstrates potential generalizability to other extreme environments through transfer learning with minimal fine-tuning.

Keywords

Seismic denoising / Edge computing / Transformer / Convolutional neural network / Mixed-precision quantization / Desert environment / Time-frequency analysis

Funding

This work was supported by the National Key R&D Program of China (grant no. 2021YFA0716800), and the CAS Project for Young Scientists in Basic Research (grant no. YSBR-020).

Conflict of interest

The authors declare they have no competing interests.

Author contributions

Conceptualization: Qingfeng Xue Formal analysis: Zesheng Yang, Qingfeng Xue, Tao Wang, Xiaoning Wang Funding acquisition: Qingfeng Xue Investigation: Zesheng Yang, Qingfeng Xue, Tao Wang Methodology: Zesheng Yang, Qingfeng Xue Validation: Qingfeng Xue Visualization: Zesheng Yang, Qingfeng Xue, Xiaoning Wang Writing–original draft: Zesheng Yang Writing–review & editing: Qingfeng Xue

Availability of data

Data collected through the research presented in the paper are available upon request to the corresponding author.

References

[1] Liu G , Chen X , Li J , Du J , Song J. Seismic noise attenuation using nonstationary polynomial fitting. Appl Geophys. 2011; 8(1): 18-26. doi: 10.1007/s11770—010—0244—2

[2] Wu S , Wang Y , Di Z , Chang X. Random noise attenuation by 3D Multi—directional vector median filter. J Appl Geophys. 2018; 159: 277-284. doi: 10.1016/j.jappgeo.2018.09.021

[3] Mousavi SM , Langston CA. Hybrid Seismic Denoising Using Higher‐Order Statistics and Improved Wavelet Block Thresholding. B Seismol Soc Am. 2016; 106(4): 1380-1393. doi: 10.1785/0120150345

[4] Yu Z , Abma R , Etgen J , Sullivan C. Attenuation of noise and simultaneous source interference using wavelet denoising. Geophysics. 2017; 82(3): V179-V190. doi: 10.1190/geo2016—0240.1

[5] Shao J , Wang Y , Liang X , Xue Qi , Liang E , Shi S. Ji yu luan sheng wang luo de ren gong zhen yuan fen bu shi guang xian chuan gan shu ju zao sheng ya zhi [Siamese network based noise elimination of artificial seismic data recorded by distributed fiber—optic acoustic sensing] . Chin J Geophys. 2022; 65(9): 3599-3609. [In Chinese]. doi: 10.6038/cjg2022P0919

[6] Li Y , Yu W , Zhang C , Yang B. Low—frequency noise suppression for desert seismic data based on a wide inference network. J Geophys Eng. 2019; 16(4): 801-810. doi: 10.1093/jge/gxz051

[7] 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

[8] Zhong T , Ye Y. MFIEN: Multi—scale feature interactive enhancement network for seismic data denoising in desert areas. Sci Rep. 2025; 15: 3979. doi: 10.1038/s41598—025—87481—y

[9] LeCun Y , Bengio Y , Hinton G. Deep learning. Nature. 2015; 521(7553): 436-444. doi: 10.1038/nature14539

[10] Mousavi SM , Beroza GC. Deep—learning seismology. Science . 2022; 377(6607): eabm4470. doi: 10.1126/science.abm4470

[11] 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

[12] Mousavi SM , Ellsworth WL , Zhu W , Chuang LY , Beroza GC. Earthquake transformer—an attentive deep—learning model for simultaneous earthquake detection and phase picking. Nat Commun. 2020; 11: 3952. doi: 10.1038/s41467—020—17591—w

[13] Zheng Y , Wang Y , Liang X , et al. A deep learning approach for signal identification in the fluid injection process during hydraulic fracturing using distributed acoustic sensing data. Front Earth Sci. 2022; 10: 999530. doi: 10.3389/feart.2022.999530

[14] Zheng Y , Wang Y. Ground—penetrating radar wavefield simulation via physics—informed neural network solver. Geophysics . 2023; 88(2): KS47-KS57. doi: 10.1190/geo2022—0293.1

[15] Wu S , Wang Y , Liang X. Joint denoising and classification network: Application to microseismic event detection in hydraulic fracturing distributed acoustic sensing monitoring. Geophysics. 2023; 88(4): L53-L63. doi: 10.1190/geo2022—0296.1

[16] Li S , Yang X , Cao A , et al. SeisT: A foundational deep learning model for earthquake monitoring tasks. IEEE Trans Geosci Remote Sens. 2024; 62: 1-15. doi: 10.1109/TGRS.2024.3371503

[17] Shao J , Wang Y , Yao Y , Wu S , Xue Q , Chang X. Simultaneous denoising of multicomponent microseismic data by joint sparse representation with dictionary learning. Geophysics. 2019; 84(5): KS155-KS172. doi: 10.1190/geo2018—0512.1

[18] Zhu W , Mousavi SM , Beroza GC. Seismic Signal Denoising and Decomposition Using Deep Neural Networks. IEEE Trans Geosci Remote Sens. 2019; 57(11): 9476-9488. doi: 10.1109/TGRS.2019.2926772

[19] Quinones L , Tibi R. Denoising Seismic Waveforms Using a Wavelet—Transform—Based Machine—Learning Method. B Seismol Soc Am. 2024; 114(4): 1777-1788. doi: 10.1785/0120230304

[20] Dantas PV , Sabino Da Silva W , Cordeiro LC , Carvalho CB. A comprehensive review of model compression techniques in machine learning. Appl Intell. 2024; 54(22): 11804-11844. doi: 10.1007/s10489—024—05747—w

[21] Ngo D , Park HC , Kang B. Edge Intelligence: A Review of Deep Neural Network Inference in Resource—Limited Environments. Electronics. 2025; 14(12): 2495. doi: 10.3390/electronics14122495

[22] Ross ZE , Meier MA , Hauksson E , Heaton TH. Generalized Seismic Phase Detection with Deep Learning. B Seismol Soc Am. 2018; 108(5A): 2894-2901. doi: 10.1785/0120180080

[23] Perol T , Gharbi M , Denolle M. Convolutional neural network for earthquake detection and location. Sci Adv. 2018; 4(2): e1700578. doi: 10.1126/sciadv.1700578

[24] Vaswani A , Shazeer N , Parmar N , et al. Attention Is All You Need. arXiv. Preprint posted online 2017. doi: 10.48550/arXiv.1706.03762

[25] Seemakhupt K , Liu S , Khan S. EdgeRAG: Online—Indexed RAG for Edge Devices. arXiv . Preprint posted online 2024. doi: 10.48550/arXiv.2412.21023

[26] Zhang F , Zhang C , Guan J , et al. Breaking the Edge: Enabling Efficient Neural Network Inference on Integrated Edge Devices. IEEE Trans Cloud Comput. 2025; 13(2): 694-710. doi: 10.1109/TCC.2025.3559346

[27] Jacob B , Kligys S , Chen B , et al. Quantization and Training of Neural Networks for Efficient Integer—Arithmetic—Only Inference. In: Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition; June 18—23 2018; Salt Lake City, UT, USA; 2018: 2704-2713. doi: 10.1109/cvpr.2018.00286

[28] Howard AG , Zhu M , et al. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. arXiv. Preprint posted online 2017. doi: 10.48550/arXiv.1704.04861

[29] Ronneberger O , Fischer P , Brox T. U—Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab N, Hornegger J, Wells WM, Frangi AF, eds. Proceedings of the Medical Image Computing and Computer—Assisted Intervention — MICCAI 2015. Vol 9351. Lecture Notes in Computer Science. MICCAI 2015; October 5—9 2015; Munich, Germany: Springer International Publishing; 2015: 234-241. doi: 10.1007/978—3—319—24574—4_28

[30] Ioffe S , Szegedy C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. In: Bach F, Blei D, eds. Proceedings of the 32nd International Conference on Machine Learning. Volume 37: International Conference on Machine Learning; July 7—9 2015; Lille, France; 2015: 448-456. https://proceedings.mlr.press/v37/ioffe15.html

[31] He K , Zhang X , Ren S , Sun J. Deep Residual Learning for Image Recognition. In: Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). June 27—30 2016; Las Vegas, NV, USA; 2016: 770-778. doi: 10.1109/cvpr.2016.90

[32] Zhou P , Xie X , Lin Z , Yan S. Towards Understanding Convergence and Generalization of AdamW. IEEE Trans Pattern Anal Mach Intell . 2024; 46(9): 6486-6493. doi: 10.1109/TPAMI.2024.3382294

[33] Micikevicius P , Narang S , Alben J , et al. Mixed Precision Training. In: Proceedings of the 2018 International Conference on Learning Representations. 6th International Conference on Learning Representations; April 30—May 3 2018; Vancouver, Canada; 2018. https://openreview.net/forum?id=r1gs9JgRZ

[34] Allen RV. Automatic earthquake recognition and timing from single traces. B Seismol Soc Am. 1978; 68(5): 1521-1532. doi: 10.1785/BSSA0680051521

[35] Saad OM , Chen Y. Deep denoising autoencoder for seismic random noise attenuation. Geophysics. 2020; 85(4): V367-V376. doi: 10.1190/geo2019—0468.1

[36] Yang L , Liu X , Zhu W , Zhao L , Beroza GC. Toward improved urban earthquake monitoring through deep—learning—based noise suppression. Sci Adv . 2022; 8(15): eabl3564. doi: 10.1126/sciadv.abl3564

[37] Yin J , Denolle MA , He B. A multitask encoder—decoder to separate earthquake and ambient noise signal in seismograms. Geophys J Int. 2022; 231(3): 1806-1822. doi: 10.1093/gji/ggac290

[38] Lomax A , Michelini A , Curtis A. Earthquake Location, Direct, Global—Search Methods. In: Meyers R, ed. Encyclopedia of Complexity and Systems Science. New York, NY, USA: Springer New York; 2009: 2449-2473. doi: 10.1007/978—0—387—30440—3_150

[39] Lapins S , Butcher A , Kendall JM, et al. DAS—N2N: machine learning distributed acoustic sensing (DAS) signal denoising without clean data. Geophys J Int. 2023; 236(2): 1026-1041. doi: 10.1093/gji/ggad460

[40] Fernández‐Carabantes J , Titos M , D’Auria L , García J , García L , Benítez C. RNN‐DAS: A New Deep Learning Approach for Detection and Real‐Time Monitoring of Volcano‐Tectonic Events Using Distributed Acoustic Sensing. JGR Solid Earth. 2025; 130(9): e2025JB031756. doi: 10.1029/2025JB031756