| CPC G01V 1/302 (2013.01) [G01V 1/307 (2013.01)] | 14 Claims |
|
10. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor of a computing system, cause the computing system to perform operations, the operations comprising:
receiving one or more first seismic surveys, wherein each of the one or more first seismic surveys comprises a set of two-dimensional (2D) lines or a three-dimensional (3D) volume, wherein each of the one or more first seismic surveys comprises a plurality of first seismic slices, and wherein each of the first seismic slices comprises a plurality of tiles;
normalizing seismic amplitudes of the tiles to be within a range from −1 to 1;
randomly shuffling the tiles, after normalizing the seismic amplitudes, to produce shuffled tiles;
reconstructing the seismic amplitudes of the shuffled tiles using an auto-encoder;
calculating a loss between the seismic amplitudes and the reconstructed seismic amplitudes using the auto-encoder, wherein the loss is calculated as a mean square error (MSE) value;
training the auto-encoder to reduce the MSE value to produce a trained auto-encoder;
receiving a second seismic survey, wherein the second seismic survey comprises a set of 2D lines or a 3D volume, wherein the second seismic survey comprises a plurality of second seismic slices;
normalizing seismic amplitudes of the second seismic slices to be within the range from −1 to 1;
converting the second seismic slices, after normalizing the seismic amplitudes of the second seismic slices, into an embedding using the trained auto-encoder or a pre-trained segmentation task-specific model, wherein the second seismic slices are converted using an encoder of the trained auto-encoder or the pre-trained segmentation task-specific model, wherein the embedding comprises one or more vectors, and wherein each of the one or more vectors comprises more than 3 dimensions;
applying a clustering algorithm to the embedding to provide a clustered embedding;
reducing a number of dimensions of the clustered embedding to 2D or 3D to produce a 2D or 3D clustered embedding;
generating a plot of the 2D or 3D clustered embedding;
identifying one or more of the second seismic slices in the clustered embedding in the plot to produce one or more identified slices;
labeling the one or more identified slices to produce one or more labeled slices; and
training a model to perform a downstream task based at least partially upon the one or more labeled slices.
|