The construction of subsurface velocity models and reservoir characterization depend heavily on the resolution of seismic interpretation. In the field of seismic exploration, combining multiple stacks, e.g. multi-angle, multi-azimuth, multi-frequency, of seismic data is becoming more and more common as a way to improve resolution. With the help of these stacks, seismic data at different offset angles can be processed, revealing more details about underlying structures and improving the imaging of intricate geological features. Delineating faults through deep learning becomes an important step in building subsurface structures. When deep learning fault prediction is used on multi-offset-angle stacks, it can help with seismic interpretation by displaying distinct fault features along each offset-angle stack. It’s still unclear, though, how to combine the outcomes of each prediction to produce the ultimate “best-of-all” output. In this abstract, we use the convolutional network to analyze each predicted fault in latent space and then combine them based on frequency analysis. The final output will mitigate break faults and compile the most dependable faults from each angle stack after combining. Combining multi-frequency or multi-azimuth faults is another application for this technique.

This content is only available via PDF.
You can access this article if you purchase or spend a download.