Skip to main contentSkip to main content
Stage 2

3D U-Net Noise Attenuation

Suppress acquisition noise — surface waves, multiples, random ambient noise — while preserving fine structural detail at every reflector horizon.

The Problem

Marine and land seismic surveys are contaminated by shot-generated noise, surface-wave energy (ground roll), multiple reflections, and ambient low-frequency noise. Conventional bandpass filtering removes noise but also attenuates the signal, blurring horizons and reducing the detectability of thin beds and DHI anomalies.

The Solution

A 3D U-Net learns a data-adaptive noise model directly from examples, separating coherent signal from incoherent noise in the full 3D spatial context. Skip connections preserve high-frequency structural detail. Streaming chunk processing ensures the approach scales to full-field acquisition volumes without GPU memory constraints.

Technical Specifications

Implementation details

ParameterValueNotes
Architecture3D U-NetRonneberger, Fischer & Brox (2015). Encoder-decoder with skip connections for full-resolution output. DOI
Weight InitializationKaiming / HeHe et al. (2015). Variance-preserving initialization for ReLU activations, prevents vanishing gradients. DOI
RuntimeONNX + INT8Jacob et al. (2018). Post-training INT8 quantization with multi-EP fallback: CUDA → DirectML → CPU. DOI
Processing ModeStreaming 64³ chunksTukey (1967). 8-sample cosine taper overlap eliminates block-boundary artefacts during reassembly.
Peak Memory< 8 GB for 50 GB volumeAIMD backpressure controller (Jacobson 1988) dynamically regulates chunk concurrency under memory pressure.
Normalization Methods7 adaptive methodsnone · peak_absolute · zscore · min_max · robust_zscore (MAD, Rousseeuw-Croux 1993) · RMS (Sheriff-Geldart 1995) · percentile clip (Dramsch 2020).

Quality Gates

Automated acceptance criteria

No result advances to the next pipeline stage unless all gating metrics are satisfied. Failures trigger automatic reprocessing with adjusted hyperparameters.

PSNR≥ 25 dB

Peak signal-to-noise ratio on held-out test traces. Failures trigger automatic reprocessing.

SSIM≥ 0.85

Wang et al. (2004). Windowed 11×11 Gaussian σ=1.5, luminance/contrast/structure decomposition.

MS-SSIMInformational

Wang, Simoncelli & Bovik (2003). Multi-scale SSIM computed at 5 spatial resolutions with power-weighted pooling.

Benchmarks

Synthetic dataset results

Evaluated on publicly available synthetic seismic benchmarks with additive Gaussian noise at multiple SNR levels.

DatasetSNR InputSNR OutputSSIMNotes
SEAM Phase I12 dB28.4 dB0.91Salt-body geometry preserved
Marmousi210 dB26.8 dB0.88Complex fold structure intact
Overthrust14 dB29.1 dB0.93Thrust fault reflectors retained

References

Peer-reviewed foundations

  1. 1Ronneberger, O., Fischer, P. & Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. MICCAI. DOI
  2. 2He, K., Zhang, X., Ren, S. & Sun, J. (2015). Delving Deep into Rectifiers. ICCV. DOI
  3. 3Jacob, B. et al. (2018). Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference. CVPR. DOI
  4. 4Wang, Z., Bovik, A. C., Sheikh, H. R. & Simoncelli, E. P. (2004). Image Quality Assessment: From Error Visibility to Structural Similarity. IEEE TIP 13(4). DOI
  5. 5Wang, Z., Simoncelli, E. P. & Bovik, A. C. (2003). Multi-scale Structural Similarity for Image Quality Assessment. Asilomar. DOI
  6. 6Tukey, J. W. (1967). An Introduction to the Calculations of Numerical Spectrum Analysis. Spectral Analysis of Time Series.
  7. 7Rousseeuw, P. J. & Croux, C. (1993). Alternatives to the Median Absolute Deviation. JASA 88(424). DOI
  8. 8Dramsch, J. S. (2020). 70 Years of Machine Learning in Geoscience in Review. Advances in Geophysics 61. DOI
Stage 2 — Denoising

See denoising on your data

Upload a SEG-Y file and watch the 3D U-Net suppress noise while preserving structural detail in under a minute.