This is award #0430208 from the Division of Information and Intelligents Systems, National Science Foundation. 

The goal of this project is to design and develop a methodology for classification and characterization of topographical features on Mars. Major tools for studying the Martian surface are geomorphic mapping and geologic mapping. The standard approach to perform these mappings is through a manual interpretation of images. This laborious approach severely limits the number of Martian sites amenable to study as modern spacecraft produces about 10 terabytes of data from a single instrument onboard. Our research aims at automating geomorphic mapping.  Different topographical variables are fused into a multi-layer data structure. Each pixel in a site carries an array of local and regional topographic information. This information is used for quantitative characterization and comparison of different topographic formations based on statistics of their constituent pixels. The results can be conveniently visualized by means of thematic maps of topography. This methodology has a potential to become a powerful investigative tool with a wide range of applications. 

We have found that digital topography data is superior to images as dataset for landform identification. Martian landforms, such as, for example, craters present formidable challenge for any identification algorithm because of the diversity of their sizes and morphologies. We have demonstrated that many of these challenges can be overcome by
using a Digital Elevation Model (DEM) data instead of imagery data. Our study resulted in an algorithm that detects fresh craters with detection percentage of 92%. We have also found that DEMs are ideal dataset for automatic mapping of different Martian landforms, valley networks. Valley networks bear some resemblance to terrestrial drainage systems,
but terrestrial mapping methods cannot be applied because of peculiarities of Martian surface. We have developed an original mapping algorithm and demonstrated its accuracy and functionality by extensively testing it against existing manual maps. We continue our effort to divide a Martian landscape into constituent, recognizable landforms. We have found that image segmentation techniques can be applied to topographic data as well. We have introduced a notion of topography object – a spatially extended area of topographic
homogeneity, and used variety of segmentation techniques to segment a landscape into a set of topography objects. These objects are classified into classes corresponding to recognizable landforms. The result is division of landscape into constituent landforms. 

T. F. Stepinski, M. P. Mendenhall, and B. D. Bue (2007) Machine Cataloging of Impact Craters on Mars. submitted to Icarus.

S. Ghosh, T. F. Stepinski, and R. Vilalta (2007) A Framework for Automatic Annotation of Planetary Surfaces with Geomorphic Labels, submitted to Machine Learning Journal

R. Vilalta, A. Bagherjeiran, T. F. Stepinski, and S. Ghosh, (2007) Semi-Local Classification Using Mixtures of Gaussians Applied to the Identification of Mars Landforms, 12th Iberoamerican Congress on Pattern Recognition, submitted.

T. F. Stepinski, S. Ghosh, and R. Vilalta (2007) Machine Learning for Automatic Mapping of Planetary Surfaces. To appear in proceedings of Nineteenth Innovative Applications of Artificial Intelligence Conference, July 24-26, 2007, Vancouver,
British Columbia. preprint

B. I. Molloy and T.F. Stepinski (2007) Automated Mapping of Valley Networks on Mars. Computers and Geoscience, 33, p728-738. paper    preprint

B. D. Bue and T.F. Stepinski (2007) Machine Detection of Martian Impact Craters from Digital Topography Data. IEEE Transactions on Geoscience and Remote Sensing, 45(1), p.265-274.  paper    preprint

R, Vilalta, T.F. Stepinski, and M. Achari (2007) An Efficient Approach to External Cluster Assessment with an Application to Martian Topography. Data Mining and Knowledge Discovery, 14(1), p.1-23.  paper    preprint

T. F. Stepinski, S. Ghosh, and R. Vilalta (2006) Automatic Recognition of Landorms on Mars Using Terrain Segmentation and Classification. In Lecture Notes in Artificial Intelligence, 4265, p 255-266.  paper    preprint

B. D. Bue and T.F. Stepinski (2006) Automated Classification of Landforms on Mars. Computers and Geoscience, 32, p.604-614.  paper    preprint

T. F. Stepinski, W. Luo, and Y. Qi (2007) Precision Mapping of Valley Networks in Margaritifer Sinus, Mars. In Lunar and Planetary Science XXXVIII, Abstract # 1205.  pdf

Ghosh, S., T. F. Stepinski, and R. Vilalta (2007) Automatic Mapping of Martian Landforms Using Segmentation-based Classification . In Lunar and Planetary Science XXXVIII, Abstract # 1200.  pdf

T. F. Stepinski, M. P. Mendenhall, and B. D. Bue (2007) Robust Automated Identification of Martian Impact Craters. In Lunar and Planetary Science XXXVIII, Abstract # 1202.  pdf

I. Molloy and T.F. Stepinski (2006) Automated Mapping of Valley Networks on Mars. In Lunar and Planetary Science XXXVII, Abstract # 1743. pdf 

B. D. Bue and T.F. Stepinski (2006) Machine Detection of Martian Craters from Digital Topography. In Lunar and Planetary Science XXXVII, Abstract # 1178. pdf

T.F. Stepinski and R. Vilalta (2005) Digital Topography Models for Martian Surfaces. IEEE Geoscience and Remote Sensing Letters, 2(3), p.260-264.  paper    preprint

B. D. Bue and T. F. Stepinski (2005) Automated Classification of Landforms in Terra Cimmeria, Mars. In Lunar and Planetary Science XXXVI, Abstract # 1195.  pdf