Identification of volcanic rootless cones and impact craters using artificial neural networks

Details Identification of volcanic rootless cones and impact craters using artificial neural networks Identification of volcanic rootless cones and impact craters using artificial neural networks

May 2004

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2004agusm.p33d..09h&link_type=abstract

American Geophysical Union, Spring Meeting 2004, abstract #P33D-09

Mathematics

Logic

3210 Modeling, 5420 Impact Phenomena (Includes Cratering), 5464 Remote Sensing, 5480 Volcanism (8450)

Scientific paper

Volcanic rootless cones -- conical mounds of scoria, spatter and excavated sediment -- form where lava interacts with underlying water-saturated sediment or ground ice to generate phreatomagmatic explosions. Terrestrial rootless cones occur in groups of tens to hundreds with individual cones ranging 2-40 m in height and 5-450 m in basal diameter. Summit craters are 0.25-0.65 of basal cone width and slopes concave to convex. Possible rootless cones identified in Mars satellite imagery are morphologically similar to terrestrial examples, but have basal diameters of 30-1000 m. Cone morphology depends largely on availability of water during lava flow emplacement. The structure and spatial distribution of rootless cones can, therefore, provide information about regolith characteristics and water abundance. As a first step toward automated identification of rootless cones in narrow angle Mars Orbiter Camera (MOC) imagery, we develop and test artificial neural network (ANN) classification algorithms using a synthetic test set. Plan-view similarity of impact craters and rootless cones with large summit craters, in addition to mantling and erosion, pose major challenges to automated recognition. Training sets include cone and crater images of varying size and complexity with normally distributed random noise added to pixel values. To simulate natural clustering of rootless cones, we model synthetic terrains by sequentially placing features drawn from the training set. The probability of cone placement at any location varies as a function of distance from existing cones. An annulus of zero probability surrounds each cone, corresponding to the region of water depletion associated with the cone-forming phreatomagmatic explosion. Probability then rises to a maximum directly beyond the depletion zone and tapers with distance to a preset background value. Backpropagating, self-organizing map, and learning vector quantization ANNs adapted to the training set are tested using synthetic terrains to assess classification accuracy. Preliminary results indicate unsupervised ANNs reliably identify and distinguish cones and craters in synthetic terrains. Classification accuracy begins to degenerate at 12% noise and at 16% consistent misclassifications occur. To test classifications in natural terrains, ongoing work applies neural networks to MOC imagery containing previously identified volcanic rootless cones. Once tested, ANNs will be used to locate additional examples in other regions.

Affiliated with

Also associated with

No associations

Rating

Identification of volcanic rootless cones and impact craters using artificial neural networks does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Identification of volcanic rootless cones and impact craters using artificial neural networks, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Identification of volcanic rootless cones and impact craters using artificial neural networks will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-1177807