Multi-blind-trace deep learning with a hybrid loss for attenuation of trice-wise noise

M.M. Abedi, D. Pardo, Tariq Ali Alkhalifah

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Trace-wise coherent noise exists in seismic data due to various reasons. We propose a self-supervised deep learning method to attenuate any type of trace-wise noise without having a clean label or noise characteristics. The proposed method is an enhanced version of blind-trace denoising. We modify the masking and calculation of loss so that the designed network reconstructs the noisy traces from the clean ones and ignores the noisy traces when reproducing the clean traces. We explain a step-by-step implementation of our method and show its application on a real deep-water dataset. The proposed method decreases the signal leakage and improves the reconstruction accuracy at and close to noisy traces.
Original languageEnglish (US)
Title of host publication84th EAGE Annual Conference & Exhibition
PublisherEuropean Association of Geoscientists & Engineers
StatePublished - 2023

Bibliographical note

KAUST Repository Item: Exported on 2023-05-29
Acknowledgements: This research is supported by the Spanish Ministry of Science and Innovation projects with references TED2021-132783B-I00, PID2019-108111RB-I00 (FEDER/AEI), the “BCAM Severo Ochoa” accreditation of excellence CEX2021-001142-S / MICIN / AEI / 10.13039/501100011033, and the Basque Government through BERC 2022-2025 program.


Dive into the research topics of 'Multi-blind-trace deep learning with a hybrid loss for attenuation of trice-wise noise'. Together they form a unique fingerprint.

Cite this