Strong noise typically interferes with seismic exploration. Under such conditions, most existing methods cannot achieve satisfactory results. Recently, methods based on diffusion models have been applied to process the strong noise in seismic data. However, the computational efficiency of diffusion-based methods is considerably lower than that of conventional methods. To address this issue, we develop a diffusion model based on the Bayesian equation and deep learning and improve its reverse equation for higher efficiency. In addition, we develop a new normalization method and an adaptive method for estimating relevant parameters. Through various improvements, our method achieves significantly better noise attenuation performance than the benchmark methods on synthetic and field data sets, while also achieving a several-fold increase in computational speed. We use transfer learning to demonstrate the robustness of our method on open-source synthetic and field data sets. Finally, we open source the code to promote the development of high precision and efficient seismic exploration methods.

You do not have access to this content, please speak to your institutional administrator if you feel you should have access.