EUVIMEDEUVIMED entdecken
Crossref

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

Zhenjing Yao, Lan Qin, Meng Lu, Jingyi Zhu, Chenyu Zhang

Journal of Seismic Exploration · 19.11.2026 · 026290142

DOI: 10.36922/jse026290142 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Seismic Exploration, online erschienen am 19.11.2026. Verfasst von Zhenjing Yao und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

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.

Zitieren

APA

Yao, Z., Qin, L., Lu, M., Zhu, J., & Zhang, C. (2026). TSU-former seismic data denoising method based on spatiotemporal attention mechanism. Journal of Seismic Exploration, 0(0), 026290142. https://doi.org/10.36922/jse026290142

Vancouver

Yao Z, Qin L, Lu M, Zhu J, Zhang C. TSU-former seismic data denoising method based on spatiotemporal attention mechanism. Journal of Seismic Exploration. 2026;0(0):026290142. doi:10.36922/jse026290142

BibTeX
@article{Yao2026,
  author = {Yao, Zhenjing and Qin, Lan and Lu, Meng and Zhu, Jingyi and Zhang, Chenyu},
  title = {TSU-former seismic data denoising method based on spatiotemporal attention mechanism},
  journal = {Journal of Seismic Exploration},
  year = {2026},
  volume = {0},
  number = {0},
  pages = {026290142},
  doi = {10.36922/jse026290142}
}
Kostenlos starten →