Answers are grounded in this paper and cite their sources.
Marine seismic interference, produced by external vessels, is a common type of noise in marine seismic data. This noise manifests as hyperbolic or linear characteristics in the shot domain, with its spectral range overlapping with that of the desired signal. Many traditional denoising methods have been proposed for the attenuation of seismic interference based on its incoherence in other domains. However, these methods may face challenges such as the degradation of desired signals. In recent years, geophysicists have begun exploring the development and applications of foundation models, which are large-scale pretrained models designed to learn general representations from data, within the geophysical domain. In this study, we introduce a foundation-model aided deep-learning (FAD) workflow for marine seismic interference attenuation. Our method utilizes a vision foundation model, the Segment Anything Model (SAM), for noise data extraction and employs the score-based diffusion model for data augmentation. Diffusion models represent state-of-the-art generative models capable of producing a multitude of data samples by learning the data distribution. The interference noise used for training is extracted before the onset of first arrivals using the SAM-based method, and the score-based diffusion model is utilized to generate diverse noise data. Subsequently, a ResUNet combined with efficient attention modules is trained to suppress interference noise. To validate the effectiveness of our proposed method, we compare it with a state-of-the-art τ-p transform-based seismic interference attenuation method and the traditional f-k approach, and analyze the impact of different output strategies, data augmentation, and network architectures on the denoising results. Experimental results demonstrate that our method offers competitive performance in noise suppression, computation time, and preservation of desired signals. These results highlight the potential of FAD approaches in advancing geophysical data processing.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: