Machine Learning workflows for seismic data interpolation - part 1 of 3 In the coming 3 weeks we will be talking about Machine Learning workflows in OpendTect for seismic data i...
Machine Learning workflows for seismic data interpolation - part 1 of 3
In the coming 3 weeks we will be talking about Machine Learning workflows in OpendTect for seismic data interpolation. On the 28th of January at 4 pm Central European Time, we will conclude the series with a free webinar. You can sign up by clicking on this link: https://lnkd.in/dHUgGkD
In this first post we show the generic nature of a U-Net interpolator. A 3D U-Net (128x128x64 samples) learned to interpolate seismic traces from examples extracted from an onshore data set over Delft. We apply the trained model to F3, an offshore dataset with different acquisition, processing and geology, located some 350 km to the North. As in Delft, we randomly blank approx. 33% of the input traces. The trained U-Net is applied “AS IS”. Next, a whole-trace RMS scaling is applied. The slider shows an in-line from F3 with: randomly blanked traces (input), RMS-scaled U-Net prediction (result), and original data (ground truth).
#OpendTect #MachineLearning
