Seismic Image Denoising With A Physics-Constrained Deep Image Prior

TL;DR

This paper presents a novel unsupervised deep learning approach using a physics-constrained Deep Image Prior to denoise seismic images by refining migration velocity models, effectively reducing artifacts and noise.

Contribution

It introduces a new method that optimizes network weights for seismic data denoising, leveraging physical constraints to improve over standard DIP techniques.

Findings

Significant noise and artifact reduction in synthetic and real seismic data.

Improved seismic image clarity with minimal computational overhead.

Outperforms standard DIP methods in seismic denoising tasks.

Abstract

Seismic images often contain both coherent and random artifacts which complicate their interpretation. To mitigate these artifacts, we introduce a novel unsupervised deep-learning method based on Deep Image Prior (DIP) which uses convolutional neural networks. Our approach optimizes the network weights to refine the migration velocity model, rather than the seismic image, effectively isolating meaningful image features from noise and artifacts. We apply this method to synthetic and real seismic data, demonstrating significant improvements over standard DIP techniques with minimal computational overhead.