Multi-scale dual-path attention network for seismic background noise attenuation.

Area of Science:

• Geophysics
• Seismic data processing
• Deep learning applications

Background:

• Background noise in seismic records hinders effective reflection event extraction, especially in complex environments like deserts.
• Non-Gaussian and nonlinear noise characteristics complicate traditional denoising, impacting seismic inversion and migration accuracy.

Purpose of the Study:

• To propose an advanced deep learning network, the Multi-scale Dual-path Attention Network (MSDPA-Net), for enhanced seismic noise reduction.
• To address limitations of single-scale feature extraction in existing deep learning models for seismic data.

Main Methods:

• MSDPA-Net utilizes a multi-scale strategy for initial feature extraction.
• A dual-path attention module is employed to differentiate between seismic signals and noise.
• Feature interaction and a reconstruction module facilitate information fusion and data restoration.

Main Results:

• MSDPA-Net demonstrated superior performance in suppressing complex seismic noise on both simulated and field data.
• The proposed network outperformed traditional denoising algorithms and standard deep learning models.
• Effective leveraging of multi-scale features significantly improved denoising effectiveness.

Conclusions:

• MSDPA-Net offers a powerful solution for seismic noise reduction in complex geological settings.
• The network's multi-scale and attention mechanisms are key to its enhanced denoising capabilities.
• This deep learning approach has the potential to improve the accuracy of seismic data interpretation and exploration.