Seismic image segmentation for detection of salt geobodies using multi-scale attributes and unsupervised classifier

in: 2019 Ring Meeting, ASGA

Abstract

The manual interpretation of seismic data is a time-consuming process and automatisation is more efficient to extract information from various types of data. We propose a work ow to semi-automatically segment seismic data into regions: salt, sediments and "uncertain". The latter contains the transition between sediments and the salt bodies. This region is created to avoid over-interpretation and localize interpretation uncertainties. To increase the robustness of the method, we combine multi-attribute classification and multi-scale representation. We choose to handle seismic attributes at coarser scales which are closer to those of the salt domes that we want to interpret. The creation of multi-scale segmented images implies combining them to build a single interpretation. The interscale fusion combines strengths from both large and small scale detections : robustness and detail preservation. Our workflow consists of four steps. First, we choose a set of seismic attributes whose answers are different enough to separate the three regions. Second, each attribute is computed at coarser scales using Gaussian pyramid decomposition. Third, a k-means clustering is applied at each scale using the previously chosen attributes to build a map. This map contains the probabilities for each pixel to be in each region. Finally, an image fusion, based on Markov random Fields and interscale dependencies, is performed to obtain a single segmented image at the finest scale. We have tested the workflow on several seismic datasets. The results demonstrate good match between seismic images and interpreted regions, and the contribution of multi-scale approach for noise reduction.

Download / Links

BibTeX Reference

@inproceedings{LegentilRM2019,
 abstract = { The manual interpretation of seismic data is a time-consuming process and automatisation is more efficient to extract information from various types of data. We propose a work ow to semi-automatically segment seismic data into regions: salt, sediments and "uncertain". The latter contains the transition between sediments and the salt bodies. This region is created to avoid over-interpretation and localize interpretation uncertainties. To increase the robustness of the method, we combine multi-attribute classification and multi-scale representation. We choose to handle seismic attributes at coarser scales which are closer to those of the salt domes that we want to interpret. The creation of multi-scale segmented images implies combining them to build a single interpretation. The interscale fusion combines strengths from both large and small scale detections : robustness and detail preservation. Our workflow consists of four steps. First, we choose a set of seismic attributes whose answers are different enough to separate the three regions. Second, each attribute is computed at coarser scales using Gaussian pyramid decomposition. Third, a k-means clustering is applied at each scale using the previously chosen attributes to build a map. This map contains the probabilities for each pixel to be in each region. Finally, an image fusion, based on Markov random Fields and interscale dependencies, is performed to obtain a single segmented image at the finest scale. We have tested the workflow on several seismic datasets. The results demonstrate good match between seismic images and interpreted regions, and the contribution of multi-scale approach for noise reduction. },
 author = { Legentil, Capucine AND Clausolles, Nicolas AND Collon, Pauline },
 booktitle = { 2019 Ring Meeting },
 publisher = { ASGA },
 title = { Seismic image segmentation for detection of salt geobodies using multi-scale attributes and unsupervised classifier },
 year = { 2019 }
}