- Conference Article
- 10.22564/19cisbgf2025.009
4D seismic inversion in the Brazilian pre-salt: Deep Learning vs. FWI
- Nov 01, 2025
- João Medeiros Araújo + 7 more +7
This study evaluates Deep Learning (DL) for time-lapse seismic inversion using Ocean Bottom Node (OBN) data, comparing it to the traditional Full Waveform Inversion Double Difference (FWIDD) method. A Convolutional Neural Network (CNN) was trained on synthetic data with velocity anomalies to predict subsurface velocity changes. The CNN demonstrated strong performance based on MSE and SSIM metrics, offering significantly faster results, as it produced predictions in seconds after a 5-minute training period, compared to FWIDD's 50-minute runtime per iteration. The DL model achieved higher resolution inversion within a known target area. Future work includes testing the CNN on unseen anomaly types and expanding its training with multiple shot locations.
Read more