Case studies:
Pattern Recognition Enables Creating Maps of Mars Landforms

Challenge

To develop a robust system for the automatic geomorphic mapping of land surfaces using a fusion of pattern recognition tools, including machine learning, computer vision, and data compression

 

Solution

  • Noesis strategy is to apply data mining algorithms to quantify points mars1or parts of the 3D seismic data to reinforce correct data interpretations. Application of algorithms to search for patterns and relationship that may exist in large databases is known as data mining.

 

  • The project uses topographic data as an input for geomorphic mapping. Utilization of topography over traditionally used imagery makes automation of mapping possible.

 

  • The project achieves the automation of geomorphic mapping by means of segmentation-based classification of landscape scenes into constituent landforms. This approach promises results that, in their appearance and content, mimic manually derived maps.

 

  • Techniques from machine learning, specifically semi-supervised learning and meta-learning, are used to minimize expemars2rt intervention and maximize efficiency during map generation.

 

  • Machine assisted quantitative comparison of generated maps on the basis of information-theoretic metrics (mutual information, information distance) extends the automation process from map generation to map analysis.

 

Results

The method is capable of correctly classifying segmenmars3ts of Mars surface with accuracy rates above 86%. Results show that inter-crater plateau, craters, and convex-crater-wall landforms are designated with high accuracy

 

Conclusions and Benefits

The results show the feasibility of using pattern recognition and machine learning to automate the process of identifying landforms on Mars. One potential application is to provide future missions an onboard application that generates geomorphic maps for the purpose of data compression. Such maps can be cheaply transmitted to Earth to assist ground controllers in decision-making.

 

Classifying landforms on Mars requires several hours of CPU time, compared to months of manual work that were needed in the past to come up with such classification. Such automation facilitates creating complete catalogs over the whole surface of the planet; catalogs can then be used to correlate the presence of specific landforms with minerals and chemicals.