News

    Machine Learning Super-sampling removes aliasing effects

    29 September 2022 Marieke van Hout-de Groot
    Machine Learning Super-sampling removes aliasing effects

    Machine Learning Super-sampling removes aliasing effects OpendTect's Machine Learning is released with a growing library of pre-trained Machine Learning models that can be used...

    Machine Learning Super-sampling removes aliasing effects

    OpendTect's Machine Learning is released with a growing library of pre-trained Machine Learning models that can be used out-of-the-box to solve similar problems on unseen datasets.

    One of the models in this library is a 3D-Unet that creates super-sampled 3D seismic volumes, i.e., volumes with half the bin-size in both directions.

    The workflow for creating a super-sampled dataset for volumes with Inline and Crossline numbers increasing with steps of 2, as is the case in the example shown here, consists of two small steps that will not take more than a few minutes:

    1. Create a Cube with alternating real seismic traces and Null traces; This is done in the Seismic Manager by copying the original volume to a new volume with output steps of 1 in the Inline and Crossline directions and “Adding Null Traces;

    2. Downloading the pre-trained 3D-Unet from the library and applying it to the volume created under 1. The original data set in this example has a bin-size of 25 x 25 m. At various places in the cube, we observe distorted seismic reflectors as a result of improper spatial sampling.

    The slider compares a crossline from the original (25 x 25 m) volume with the super-sampled (12.5 x 12.5 m) volume. Note, that although interpolation is no substitute for acquiring real data, the super-sampled dataset has drastically reduced the amount of aliased energy. Apparently, the Machine Learning model was able to reconstruct signals that were distorted in 2D section view by interpolating in 3 dimensions. Note 2, if Inlines and Crossline numbers increase with steps of 1, the data needs to be renumbered. In that case it is best to setup a new survey and renumber Inlines and Crosslines during SEGY import.

    Author dr. Paul de Groot from dGB Earth Sciences

    #OpendTect #machinelearning #seismic #geoscience #geophysics

    NewsOpendTectdGBMachine LearningSeismicReleaseMachinelearningGeoscience