Multichannel blind deconvolution assisted by deep learning

View at publisher Summarise with VIRAT
Authors: Andrés Landeta, Mauricio D. Sacchi, Amélia Novais, Jörg Dietrich Wilhelm Schleicher
Cited by: 0
References: 5

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Blind deconvolution, i.e., the simultaneous recovery of the seismic wavelet and reflectivity, is a fundamental challenge in seismic exploration. We tackle this ill-posed problem by means of a multichannel blind deconvolution approach as sisted by deep learning, where machine learning provides an initial wavelet estimate. Specifically, we train a deep convolutional network to predict the wavelet from synthetic zero offset seismic data. This predicted wavelet serves as the starting point for an minimization algorithm that alternatingly improves the reflectivity estimate and the wavelet. A well initialized wavelet can significantly enhance the inversion process, leading to accurate wavelet and reflectivity recovery. However, the algorithm may converge to an incorrect solution if the initial estimate is too far from the true wavelet. We demonstrate that a machine-learning initial wavelet improves the chances for success. This underscores the critical role of machine learning in generating a reliable wavelet initialization, ultimately improving the robustness and stability of blind deconvolution.

Fields of Study & Subjects

Seismic Imaging and Inversion Techniques Seismic Waves and Analysis Image and Signal Denoising Methods

Publication details

Journal / source
Second International Meeting for Applied Geoscience & Energy
Volume
1
Pages
135–139
Publication date
10 February 2026
Language
EN