Self-supervised training for simultaneous interpolation and deblending of blended data

View at publisher Summarise with VIRAT
Authors: Zifei Li, Shaohuan Zu, Wanting Shen, Jiale Kang, Jingxiao Wang
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

Deep learning technology has been widely applied in the processing of simultaneous source blended data, including the deblending of blended data and the interpolation of missing traces. When dealing with blended data, it is often necessary to address the issues of blending noise and missing traces. There are typically two approaches for processing such data: compressed sensing theory and deep learning. However, compressed sensing methods require extensive analysis and parameter tuning, while deep learning methods require large datasets or, in the case of self-supervised methods, often require specific adjustments to the network model. To simplify the problem of simultaneous interpolation and separation of blended data, we propose a self-supervised training method that can achieve relatively high-quality results without the need for prior analysis of the data or adjustments to the network model. We conducted experiments using a set of synthetic data and a set of real data to validate our results.

Fields of Study & Subjects

Seismic Imaging and Inversion Techniques Seismic Waves and Analysis Neural Networks and Applications

Publication details

Journal / source
Second International Meeting for Applied Geoscience & Energy
Volume
1
Pages
2170–2174
Publication date
10 February 2026
Language
EN