Fast Diffusion Model For Seismic Data Noise Attenuation

TL;DR

This paper introduces an improved diffusion model for seismic data noise attenuation that enhances noise removal effectiveness and computational efficiency, validated on synthetic and field datasets with open-source code release.

Contribution

The paper presents a novel diffusion-based seismic noise attenuation method with an improved Bayesian iteration scheme and normalization, achieving higher accuracy and speed.

Findings

Significantly better noise attenuation than benchmark methods

Several-fold increase in computational speed

Robust performance demonstrated through transfer learning

Abstract

Noise is one of the primary sources of interference in seismic exploration. Many authors have proposed various methods to remove noise from seismic data; however, in the face of strong noise conditions, satisfactory results are often not achievable. In recent years, methods based on diffusion models have been applied to the task of strong noise processing in seismic data. However, due to iterative computations, the computational efficiency of diffusion-based methods is much lower than conventional methods. To address this issue, we propose using an improved Bayesian equation for iterations, removing the stochastic terms from the computation. Additionally, we proposed a new normalization method adapted to the diffusion model. Through various improvements, on synthetic datasets and field datasets, our proposed method achieves significantly better noise attenuation effects compared to the benchmark methods, while also achieving a several-fold increase in computational speed. We employ transfer learning to demonstrate the robustness of our proposed method on open-source synthetic seismic data and validate on open-source field data sets. Finally, we open-sourced the code to promote the development of high-precision and efficient seismic exploration work.