We propose a procedure for the interpretation of horizons in seismic reflection data based on a Neural Network (NN) approach, which can be at the same time fast, accurate and able to reduce the intrinsic subjectivity of manual or control-points based methods. The training is based on a Long Short Term Memory architecture and is performed on synthetic data obtained from a convolutional model-based scheme, while the extraction step can be applied to any type of field seismic dataset. Synthetic data are contaminated with different types of noise to improve the performance of the NN in a large variety of field conditions. We tested the proposed procedure on 2-D and 3-D synthetic and field seismic datasets. We have successfully applied the procedure also to Ground Penetrating Radar data, verifying its versatility and potential. The proposed algorithm is based on a fully 1-D approach and does not require the input of any interpreter, because the necessary thresholds are automatically estimated. An added benefit is that the prediction has an associated probability, which automatically quantifies the reliability of the results.

Efficient extraction of seismic reflection with Deep Learning

Roncoroni, G.
;
Forte, E.;Bortolussi, L.;Pipan, M.
2022

Abstract

We propose a procedure for the interpretation of horizons in seismic reflection data based on a Neural Network (NN) approach, which can be at the same time fast, accurate and able to reduce the intrinsic subjectivity of manual or control-points based methods. The training is based on a Long Short Term Memory architecture and is performed on synthetic data obtained from a convolutional model-based scheme, while the extraction step can be applied to any type of field seismic dataset. Synthetic data are contaminated with different types of noise to improve the performance of the NN in a large variety of field conditions. We tested the proposed procedure on 2-D and 3-D synthetic and field seismic datasets. We have successfully applied the procedure also to Ground Penetrating Radar data, verifying its versatility and potential. The proposed algorithm is based on a fully 1-D approach and does not require the input of any interpreter, because the necessary thresholds are automatically estimated. An added benefit is that the prediction has an associated probability, which automatically quantifies the reliability of the results.
Pubblicato
https://www.sciencedirect.com/science/article/pii/S0098300422001431
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S0098300422001431-main.pdf

non disponibili

Tipologia: Documento in Versione Editoriale
Licenza: Copyright Editore
Dimensione 4.95 MB
Formato Adobe PDF
4.95 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11368/3025626
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
social impact