Seismic Interpretation analysis with deeplarning
Datasets
Ours
For Unet-3d we propose our labled dataset in masks+slices folder. It consist of images sliced horizontally from SEG-Y files from New Zeland.
Rosneft booster's challenge
Link Labled crosslines split for seismic facies analysis. Used
Netherlands Dataset
Link Contains pos-stackdata, 8 horizons and well logs of 4 wells. For the purposes of ourmachine learning tasks, the original dataset was reinterpreted,generating 9 horizons separating different seismic facies intervals.The interpreted horizons were used to generateā¼190,000 labeledimages for inlines and crosslines.
Results
Crosslines
Our Unet + Resnet model achived 0.97 (max value - 1.00) multidice metric on the validation set.