Application of Machine Learning Technology to Martian Geology

1
Application of Machine Learning Technology to
Martian Geology

• Last Years Highlights
• Calibrated TES data, registered to map base,
  retained high quality data with minimum
  atmospheric influences, constructed spectral
  database of laboratory measurements of relevant
  materials
• Applied supervised and unsupervised
  classification algorithms and spectral unmixing
  to determine mineralogical composition of study
  area
• found strong differences between abundances of
  rock-forming minerals in mountain and plain
  regions.

Impact on Science
Ruye Wang

• Explore how latest machine learning development
  can be applied to remote sensing data processing
  and analysis
• Discover strengths and limitations of various
  algorithms in terms classification of materials
  of low abundance with high noise
• Provide new computational tools to benefit
  geological study on Mars

Harvey Mudd College Ruye_Wang_at_hmc.edu
James Dohm The University of Arizona
Rebecca Castonio Jet Propulsion Lab, NASA
Project Summary

• Development of a software system for robust
  detection and classification of different types
  of rocks and minerals of geologic interest on
  Mars, using THEMIS and TES data, applied to
  analyze a region on Mars of geological
  significance.

• Plans for Next Year
• Develop Bayesian based classification algorithms
• Compare with previous classification results
• Evaluate classification algorithms in low
  abundances and high noise .