Download presentation
Presentation is loading. Please wait.
Published byAngela Evans Modified over 5 years ago
1
Image segmentation for velocity model construction and updating
Adam Halpert* Robert G. Clapp SEP-134, 13 May 2008
2
Why segmentation? Velocity model building is a “bottleneck” for large iterative imaging projects Manually picking salt boundaries is often tedious, ambiguous, and time-consuming Image segmentation offers a consistent, efficient means of delineating salt bodies Integration with velocity model building produces many benefits 1
3
Before 2
4
After 3
5
Agenda Segmentation review Migration algorithms
Effects on segmentation Velocity model construction Sediment flood, salt flood Velocity model updating Optimization of the boundary pick Segmentation Review Migration Model Construction Model Updates 4
6
Segmentation defined First applied to seismic data by Hale and Emanuel (2002, 2003): Atomic meshing Lomask (2007): Divide a seismic image into two volumes - a salt body, and the surrounding sediments Based on tracking one or more key attributes ( i.e. envelope amplitude) that can identify likely salt interfaces Ultimate goal: Produce globally accurate salt boundaries with minimal human interaction Segmentation Review Migration Model Construction Model Updates 5
7
Segmentation algorithm
Normalized cuts image segmentation (Shi and Malik, 2000) Divide a migrated seismic image into pixels For each pixel, compare values of envelope amplitude between it and a random selection of neighbors For each pair of pixels, assign a weight inversely proportional to the likelihood of a salt boundary between them From Lomask (2007) Segmentation Review Migration Model Construction Model Updates 6
8
Segmentation algorithm
Normalized cuts image segmentation (Shi and Malik, 2000) Divide a migrated seismic image into pixels A path that minimizes the sum of weights across an image is the salt boundary From Lomask (2007) Segmentation Review Migration Model Construction Model Updates 6
9
Segmentation algorithm
The expression for a vector representing the minimized cut can be expressed as a Rayleigh quotient: W: matrix containing all calculated weights D: diagonal matrix formed from W’s columns Segmentation Review Migration Model Construction Model Updates 7
10
Segmentation algorithm
This is an eigenvector/eigenvalue problem Segmentation Review Migration Model Construction Model Updates 8
11
Segmentation algorithm
The eigenvector corresponding to the second- smallest eigenvalue is used to segment the image When following the “zero contour,” positive eigenvector values are in one group (salt), negative values in the other (sediment) Segmentation Review Migration Model Construction Model Updates 9
12
Example: Image Segmentation Review Migration Model Construction Model Updates 10
13
Example: Eigenvector Segmentation Review Migration Model Construction Model Updates 11
14
Migration Segmentation operates on migrated images
Higher quality images higher quality segmentation Plane wave migration in tilted coordinates (Shan, 2008) excels at imaging steeply- dipping structures Ideal for salt-boundary delineation Segmentation Review Migration Model Construction Model Updates 12
15
Image: Regular coordinates
Segmentation Review Migration Model Construction Model Updates 13
16
Image: Tilted coordinates
Segmentation Review Migration Model Construction Model Updates 14
17
Eigenvector: Regular coordinates
Segmentation Review Migration Model Construction Model Updates 15
18
Eigenvector: Tilted coordinates
Segmentation Review Migration Model Construction Model Updates 16
19
Quantitative comparison
Regular coordinates Tilted coordinates Segmentation Review Migration Model Construction Model Updates 17
20
Image: Regular coordinates
Uncertainties Segmentation Review Migration Model Construction Model Updates 18
21
Image: Tilted coordinates
Uncertainties Segmentation Review Migration Model Construction Model Updates 19
22
Eigenvector: Regular coordinates
Uncertainties Segmentation Review Migration Model Construction Model Updates 20
23
Eigenvector: Tilted coordinates
Segmentation Review Migration Model Construction Model Updates 21
24
Quantitative comparison
Regular coordinates Tilted coordinates Segmentation Review Migration Model Construction Model Updates 22
25
Velocity model building
Common procedure: Sediment flood migration resolves salt top Flood with salt velocity below picked top boundary Salt flood migration resolves salt base Segmentation Review Migration Model Construction Model Updates 23
26
Stratigraphic velocity model
Segmentation Review Migration Model Construction Model Updates 24
27
Sediment flood velocity model
Segmentation Review Migration Model Construction Model Updates 24
28
Stratigraphic velocity model
Segmentation Review Migration Model Construction Model Updates 24
29
Sediment flood image Segmentation Review Migration Model Construction Model Updates 25
30
Sediment flood eigenvector
Segmentation Review Migration Model Construction Model Updates 26
31
Zero-contour boundary
Segmentation Review Migration Model Construction Model Updates 27
32
Salt flood image Segmentation Review Migration Model Construction Model Updates 28
33
Salt flood eigenvector
Segmentation Review Migration Model Construction Model Updates 29
34
Zero-contour boundary
Segmentation Review Migration Model Construction Model Updates 30
35
Updating velocity models
Segmentation can improve pre-existing velocity models Now, the existing model acts as a priori information for choosing salt boundaries In uncertain (grey) areas, follow (or stay away from) the boundary on the existing model Segmentation Review Migration Model Construction Model Updates 31
36
Beyond the zero-contour
Lomask’s algorithm follows a single eigenvector contour value across the entire image The “best” boundary likely follows different values of the second eigenvector in different parts of the image Pose the boundary pick as an optimization problem Segmentation Review Migration Model Construction Model Updates 32
37
Boundary optimization
This is a NONLINEAR problem Primary goal: follow the zero-contour Linearize the problem around the zero-contour boundary Address the problem via a series of fitting goals Segmentation Review Migration Model Construction Model Updates 33
38
Boundary optimization
Create a linear operator, G G is the vertical gradient of the eigenvector across the zero-contour boundary G is large when there is a sharp transition from postive to negative values, small when the transition is gradual 0 G ∆m Deviation from zero-contour boundary Segmentation Review Migration Model Construction Model Updates 34
39
Boundary optimization
Use the boundary from the existing model when the zero-contour is ambiguous Weighting operator W is large when there is great uncertainty in the eigenvector: 0 W (m-mprior) Segmentation Review Migration Model Construction Model Updates 35
40
Boundary optimization
Impose a smoothness constraint on the modeled boundary 0 A m, where A is a roughening operator, in this case the gradient Segmentation Review Migration Model Construction Model Updates 36
41
Boundary optimization
0 G ∆m 0 W (m-mprior) 0 A m Segmentation Review Migration Model Construction Model Updates 36
42
Image after salt flood migration
Segmentation Review Migration Model Construction Model Updates 37
43
Image after salt flood migration
Bottom of canyon placed too shallow Segmentation Review Migration Model Construction Model Updates 37
44
Zero-contour boundary
Still too shallow… move away from zero contour Segmentation Review Migration Model Construction Model Updates 38
45
Optimized boundary Segmentation Review Migration Model Construction Model Updates 39
46
Remigrated image Segmentation Review Migration Model Construction Model Updates 40
47
Original Image Segmentation Review Migration Model Construction Model Updates 41
48
Remigrated image Canyon imaged better, but still placed too shallow
Segmentation Review Migration Model Construction Model Updates 42
49
One more iteration Original zero contour First optimization iteration Second optimization iteration Segmentation Review Migration Model Construction Model Updates 43
50
Final remigrated image
Segmentation Review Migration Model Construction Model Updates 44
51
Previous Iteration Segmentation Review Migration Model Construction Model Updates 44
52
Final remigrated image
Segmentation Review Migration Model Construction Model Updates 44
53
Perfect velocity migration
Segmentation Review Migration Model Construction Model Updates 45
54
Real data example: Original image
Segmentation Review Migration Model Construction Model Updates 46
55
Original velocity model
Segmentation Review Migration Model Construction Model Updates 47
56
Optimized salt boundary
Segmentation Review Migration Model Construction Model Updates 48
57
Updated velocity model
Segmentation Review Migration Model Construction Model Updates 49
58
Original velocity model
Segmentation Review Migration Model Construction Model Updates 50
59
Remigrated image Segmentation Review Migration Model Construction Model Updates 51
60
Original image Segmentation Review Migration Model Construction Model Updates 52
61
Conclusions Image segmentation can accurately identify salt boundaries
Segmentation can greatly expedite the velocity model-building process by operating on sediment-flood and salt-flood images When a velocity model already exists, optimizing the algorithm’s boundary picks produces improved velocity models and migrated images 53
62
Acknowledgments We thank WesternGeco and SMAART JV for supplying the data used in these examples, and several SEP students for providing suggestions and technical guidance. 54
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.