Classification of Mars Terrain Using Multiple Data Sources
Description
Classification of Mars Terrain Using Multiple Data Sources
Alan Kraut1, David Wettergreen1
ABSTRACT. Images of Mars are being collected faster than they can be analyzed by planetary
scientists. Automatic analysis of images would enable more rapid and more consistent image
interpretation and could draft geologic maps where none yet exist. In this work we develop a
method for incorporating images from multiple instruments to classify Martian terrain into
multiple types. Each image is segmented into contiguous groups of similar pixels, called
superpixels, with an associated vector of discriminative features. We have developed and
tested several classification algorithms to associate a best class to each superpixel. These
classifiers are trained using three different manual classifications with between 2 and 6 classes.
Automatic classification accuracies of 50 to 80% are achieved in leave-one-out cross-validation
across 20 scenes using a multi-class boosting classifier.
Resources
| Name |
Format |
Description |
Link |
|
33 |
Classification of Mars Terrain Using Multiple Data Sources |
https://c3.nasa.gov/dashlink/static/media/publication/Paper_5_.pdf |