Automatic Salt Dome Detection Using U-net
Y.X. Zhang, Y. Liu, H.R. Zhang and H. Xue
Event name: 81st EAGE Conference and Exhibition 2019
Session: AI/Digitalization for Interpretation - Geological and Model Building Interpretation
Publication date: 03 June 2019
Info: Extended abstract, PDF ( 2.9Mb )
Price: € 20
Accurate salt dome detection is crucial for 3D seismic data interpretation. However, the main drawbacks of most existing detection techniques are the low positioning accuracy and quite long computing time. To solve these drawbacks, we regard the salt dome detection as an object segmentation problem and propose the U-net segmentation network and a new training strategy. We construct the neural network, make two types of salt dome labels, and use few annotated samples to automatically generate massive training data, then train the network model and test the model through unknown data. The results indicate that the proposed network can accurately describe the salt dome boundaries with small computational cost and be applicable to entire seismic volume.