Pattern Recognition and its use for easy seismic trace shape mapping

1
Pattern Recognition and its use for easy seismic
trace shape mapping ? An expose of the what,
who, why, when, how for busy geoscientists Autho
r J.D. Kerr - Landmark EAME jkerr_at_lgc.com Pre
sented by Charles Thomas
2
What ?
Is Pattern Recognition
3
Pattern recognition is the research area that
studies the operation and design of systems that
recognize patterns in data
It encloses subdisciplines like 1)
Discriminant analysis 2) Feature
extraction 3) Error estimation 4) Cluster
analysis 5) Grammatical inference and parsing
4
Pattern Recognition ? Can you recognize this
object ?
How well trained are your visual neurons ?
5
Pattern Recognition ? Can you recognize this
object YET ?
6
A pint of homebrewed stout !
7
How did you do ?
Well trained neurons can recognize classify a
pint of stout with only the most basic image!
The challenge is to create pattern recognition
software capable of approaching our brains
ability. E.g. Artificial Neural Network s/w
K nearest neighbour classification s/w
8
Who ?
Uses Pattern Recognition software
9
COMPUTATION TECHNOLOGY GROUP The Advanced
Computation Technology Group, Code B10 of the
Naval Surface Warfare Center Dahlgren Division
(NSWCDD) is active in applying pattern
recognition and image processing techniques to a
number of different domains. In addition to Naval
applications, we are particularly interested in
medical imaging.
Abstract Comparative Evaluation of Pattern
Recognition Techniques for Detection of
Microcalcifications in Mammography
Seven classifiers (linear and quadratic
classifiers, binary decision trees, a standard
backpropagation network, 2 dynamic neural
networks, and a K-nearest neighbor) are compared.
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
PATTERN RECOGNITION IN GEOSCIENCES
Geological interpretation 1) Well log
correlation. 2) Satellite land use
mapping Seismic processing 1) First
Break picking of refractions. 2) Bad
trace editing of field records. Seismic
interpretation 1) Horizon surface
interpretation. 2) Seismic trace shape
mapping.
18
Satellite land use mapping
Wavelengths
Intensities
Light reflectivity signature
Satelite recording of infrared to ultraviolet
light reflectivity enables land use maps to be
made based on reflectivity signatures.
19
Seismic trace shape mapping
Pattern Recognition
Class map 1 Parms 1, 95, 2, 25 )
Class map 2 Parms 1, 95, 1, 45 )
Trace character anomaly map
Class map 3 Parms 1, 95, 0.1, 45 )
20
Why ?
use Pattern Recognition for seismic trace shape
mapping
21
Geophysics, Vol 58 No 10 October 1993 Amplitude
map analysis using forward modeling in Sandstone
and carbonate reservoirs Dennis B Neff, Phillips
Petroleum Co
22
(No Transcript)
23
When ?
would you use Pattern Recognition for seismic
trace shape mapping
24
Seismic trace shape
Characteristic Attribute
Trace classification combines various trace
attributes
25
Trace attribute approach
Attribute Extraction
RMS amplitude or Max peak/trgh amp
Ave instantaneous freq or Peak spectral freq
Trace character anomaly map
Response phase or 1/2 Energy time
26
Trace shape approach
Pattern Recognition
Class map 1 Parms 1, 95, 2, 25 )
Class map 2 Parms 1, 95, 1, 45 )
Trace character anomaly map
Class map 3 Parms 1, 95, 0.1, 45 )
27
Seismic Trace shape mapping
5
Pattern Recognition
Class map 1
Classification using pre-defined statistical
rules
5
2
2
5
11
Rules for class membership allow 15 classes
Rules for class membership allow 25 classes
Class map 2
Trace character anomaly map
Class map 3
28
How ?
would you use Pattern Recognition for seismic
trace shape mapping
29

SUPERVISED K CLASSIFICATION
30
Satellite land use mapping
Wavelengths
Intensities
Light reflectivity signature
Satelite recording of infrared to ultraviolet
light reflectivity enables land use maps to be
made based on reflectivity signatures.
31
Class allocation
Pattern recognition using known classes
Light reflectivity signature
A concrete B grass C deep water D sand
A concrete B grass C deep water D sand
Intensities
D
SUPERVISED CLASSIFICATION for land use maps
32
Class allocation
Pattern recognition using RULES
Light reflectivity signature
Statistical RULES for allocating classes
controling merging splitting
A buildings ? B vegetation ? C water ? D
soil ?
UNSUPERVISED CLASSIFICATION for land use maps
33
Class allocation
Pattern recognition using RULES
Statistical RULES for allocating classes
controling merging splitting
UNSUPERVISED CLASSIFICATION for seismic trace
shape maps
34
ER Mapper ISOCLASS Unsupervised classification
35
LandMark StratTrac Unsupervised classification
36
2 horizon slices Points are amplitude No spatial
information, but can see amplitude clustering
Initial unclassified scattergram
37
Allocating members to 3 seed classes
38
Splitting of class B with too much deviation
39
Merging of classes A B2 with class means too
close
40
Final stable result of classes A, B1 C
41
How ?
would you interpret seismic trace shape maps
42
Original salt3d seismic data showing Top Sst to
Base Buffet horizons and Top Sst 40msec marker.
43
Pretty maps but .. what do they mean ?
44
Comparison of input horizon slices with the
output class map. ( LHS slice at 12msec RHS
best original class map )
45
Comparison of input horizon slices with the
output class map. ( LHS slice at 20msec RHS
best original class map )
46
Comparison of interval RMS amplitude with the
output class map. ( LHS RMS RHS best
original class map )
47
Comparison of an original amplitude class map
with a cosine of phase class map. ( LHS
original RHS cosine of phase )
48
Comparison of peak-gttrgh thickness with the
output class map. ( LHS thickness RHS best
cosine of phase class map )
49
Tying the Waveform Classification to the Geologic
Model Developed from the Attribute Analysis

• Channel
• Levee
• Splay
• Fan Lobe
• Basin
• Terrace
• Bars
• Other

We are looking for these 8 geological components
of our channel/fan complex
from Facies Models, 2nd edition, Walker 1984
50
Unsupervised Waveform Classification
8 classes symmetrical window /-20 ms
8 classes asymmetrical window 36 ms
Flow direction in distal Channel/Fan
Terrace
Mouth bar or Lateral Flow?
51
Before and After.
8 classes asymmetrical window 36 ms
Channel infill has a good amplitude response
52
Conclusions

• Pattern Recognition techniques promise direct
  seismic trace shape mapping and easier
  interpretation However
• Care must be taken to choose rules for class
  membership which allow a reasonable spread of
  member numbers and their statistical distribution
• As with normal attribute maps, class maps are
  sensitive to calculation window width and
  location
• Normal attributes can give better maps using
  variable width windows than constant window class
  maps

53
Pattern Recognition ? Can you recognize this
object ?
If not .then its time to go and get some
training !