Seismic data denoising by diffusion model
Ji Li, Daniel O. Trad
Diffusion models have become popular in generative modelling with neural networks,showing strong performance in tasks like denoising and super-resolution. However, most existing methods rely on large amounts of training data and work well within the samedata distribution but struggle with distribution shifts. This paper presents an unsupervised diffusion model for seismic signal denoising that requires only noisy seismic data without needing extra clean data for training. Experiments on noisy seismic data show that the proposed method is more robust to noise than traditional diffusion models and other unsupervised methods.