presentation given at the fall agu meeting san francisco...
TRANSCRIPT
Presentation given atthe Fall AGU Meeting
San Francisco, December 2006
Robust System for Identification of Martian Impact Craters
Tomasz F. StepinskiLunar and Planetary Institute
Michael P. MendenhallWashington Univ.in St. Louis
Brian D. BueJPL
Young, Amazonian surface, MDIM 2.1 image
The Catalog of Large MartianImpact Craters
(Barlow, Icarus 75, p285, 1988)
> 42,000 craters
Mars has variety of different kind of craters
Old, Noachian surface, MDIM 2.1 image
Hesperian surface, MDIM 2.1 image
Identification of craters from topographic dataADVANTAGES:• DEM provides a direct, three-dimensional representation of Martian surface.
• No problems associated with “visibility.”
• Possibility of calculating more parameters, such as crater depth.
DISADVANTAGES:• Low resolution of available data (MOLA).
THEMIS image, 100 meters/pixel
Rendering of MOLA topography, ~500 meters/pixel
Identification of craters from topographic dataSecond generation approach
Design criteria:• Robust, works well on all types of Martian surfaces.
• Fast, permitting generating a catalog of craters over the entire Martian surface.
• Scale-independent, can be applied to dataset other than MOLA.
• Simple, can be offered as a download.
“Crater finding” transform: an example
Tisia Valles, 383 X 422 pixels transform, r = 5 pixels transform, r = 10 pixels
transform, r = 20 pixels
• The C-Transform smoothes out features smaller than the characteristic size r.
• The C-Transform suppresses low-frequency components in topography, leveling out any slowly changing background gradients.
• A crater of the characteristic size creates a smooth, pronounced basin in the transform surface.
Process of finding craters for layer r = 5 pixels
(1) transform, r = 5 pixels (2) Identified depressions (3) After shape selection
Process of finding craters for layer r = 20 pixels
Final results: Tisia Valles
31 craters identified
No false positives
Some false negatives
Catalog contains:
1) Coordinates2) Radius3) Shape descriptors4) Depth
• New amazing CRATERMATC software.
• Performs C-Transform, crater cores identification, expansion, shape-based selection, all in one easy-to-use package.
• Fast, written in C++
• Produces catalog of crater candidates and images.
• Results can be used as an input to the machine learning-based selection algorithm
• Free!
• Available at cratermatic.sourceforge.net
Technical details on final selectionusing machine learning
• We use WEKA – an open source collection of machine learning algorithms for data mining tasks.• Different (Yes/No) classifier is constructed for each layer using 5 features: radius, depth, radius/depth ratio, two shape coefficients.
• Initial training set was labeled by human expert.
• Accuracy of the classifier for layer 1 is 95%, for other layer it is >85%.
• Obtained classifier is used for all unlabeled crater candidates.
Example: Terra Cimmeria #1, 1269 candidates, 734 craters
Example: Terra Cimmeria #1, details
Example: Terra Cimmeria #1, details
Example: Hesperia Planum, 554 candidates, 305 craters
Example: Hesperia Planum, details
Example: Hesperia Planum, details
Name Coordinates (lower left, long., lat.) (upper right, long., lat.)
Area (millions square km.)
# of craters(our code)
# of craters(Barlow)
Terra Cimmeria #1
(114.0, -18.42)(141.4, -7.58)
1.0 734 499
Terra Cimmeria #2
(117.4, -28.4)(145.4, -17.0)
0.98 748 571
Terra Cimmeria #3
(117.4, -38.6)(145.4, -26.6)
0.92 662 508
Terra Cimmeria #4
(117.4, -47.5)(145.4, -36.5)
0.73 300 409
Hesperia Planum
(107.1, -29.6)(118.5, -17.0)
0.44 305 136
Sinai Planum
(261.5, -29.7)(278.6, -10.3)
1.0 468 124
Identification of craters: automatic vs. manual
Automatic vs. manual: comparison of crater counts
Red - craters present in the Barlow catalogand identified by our algorithm.
Green - craters identified by our algorithmbut absent from the Barlow catalog.
Blue - craters present in the Barlow catalogbut not identified by our algorithm.
Automatic vs. manual: comparison of craters’ diameters
Comparison between:Our dataset (automatic)Barlow catalog (manual)
1370 craters in:Hesperia PlanumSinai PlanumTerra Cimmeria
Our values of diameterare systematically largerby about 15%.
Automatic vs. manual: comparison of craters’ depths
Comparison between:Our dataset (automatic)Joe Boyce dataset (manual)
144 craters inHesperia PlanumSinai Planum
Our values of depth aresystematically larger by about 30%.
Main reason for observeddiscrepancy:Difference in how craterdepth is defined.
Diameter – depth statistics: Terra Cimmeria
Diameter – depth statistics: Terra Cimmeria
Conclusions• Our DEM-based system for identification and characterization of Martian craters is robust and practical.
• We are finding 43% more craters than in the Barlow catalog. These are mostly small craters.
• There are almost no false positives.
• We have established a large training set to be used for identification of craters over the entire Martian surface.
• Our estimate of craters’ radii and depths compare well with results of manual measurements.
Future• Compile catalog of craters over the entire Mars surface.
• Apply our system to Martian datasets other than MOLA.
• Apply our system to datasets from other planets.