This project aims to provide an efficient and accurate deep learning algorithm for seismic data wavefield separation. The algorithm functions similarly to F-K filtering and Radon transform filtering and is applicable to various scenarios, including VSP up- and down-going wave separation, DAS-VSP single-component P/S separation based on apparent velocity differences, multiple-wave attenuation in CMP and CRP gathers, and linear noise attenuation. This method is suitable for tasks where the wavefields to be separated have significant apparent velocity differences.

The project is built on the PyTorch framework and includes training datasets and pre-trained models. It is open-sourced under the CC BY-SA 4.0 License, allowing use, sharing, adaptation, and redistribution with proper attribution to the author. The code is provided “as-is,” without any warranties or liability.

This project will continue to improve and expand the datasets. If you have any seismic data that you are willing to share for this purpose, please feel free to contact me.

Author: Xiaobin Li

Contact me: lixb9767@163.com