MDA GAN: Adversarial-Learning-Based 3-D Seismic Data Interpolation and Reconstruction for Complex Missing

The interpolation and reconstruction of missing traces is a crucial step in seismic data processing, moreover it also highly ill-posed problem, especially for complex cases such as high-ratio random discrete missing, continuous fault-rich or salt body surveys. These are rarely mentioned current works. To cope with cases, we propose Multi-Dimensional Adversarial GAN (MDA GAN), novel 3-D framework. It keeps anisotropy spatial continuity the after 3D using three discriminators. feature stitching module designed embedded generator to retain more information input data. Tanh cross entropy (TCE) loss derived, which provides optimal gradient make generated smoother continuous. We experimentally verified effectiveness individual components study then tested method on multiple publicly available achieves reasonable reconstructions up 95% 100 missing. In fault enriched surveys, MDA still yields promising results cases. Experimentally has been demonstrated that our better performance than other methods both simple cases.https://github.com/douyimin/MDA_GAN