HR: 1340h
AN: P43A-0911    [Abstracts]
TI: Generation and Performance of Automated Jarosite Mineral Detectors for Mars Rovers.
AU: Merrill, M D
EM: mmerrill@wesleyan.edu
AF: Dept. of Earth and Environmental Sciences, Wesleyan University, Middletown, CT 06459 United States
AU: * Bornstein, B
EM: ben.bornstein@jpl.nasa.gov
AF: Jet Propulsion Laboratory, 4800 Oak Grove Drive, Pasadena, CA 91109 United States
AU: Gilmore, M S
AF: Dept. of Earth and Environmental Sciences, Wesleyan University, Middletown, CT 06459 United States
AU: Casta\~{n}o, R
AF: Jet Propulsion Laboratory, 4800 Oak Grove Drive, Pasadena, CA 91109 United States
AU: Greenwood, J
AF: Dept. of Earth and Environmental Sciences, Wesleyan University, Middletown, CT 06459 United States
AB: Sulfate salt discoveries at the Eagle and Endurance craters in Meridiani Planum by the Mars Exploration Rover Opportunity have proven mineralogically the existence and involvement of water in Mars' past. Visible and near infrared spectrometers like the Mars Express OMEGA, the upcoming 2006 Mars Reconnaissance Orbiter CRISM and the 2009 Mars Science Laboratory Rover cameras may facilitate the identification of water-bearing salts. Increasing spectral resolution and rover mission lifetimes currently necessitate greater data compression in order to ease downlink restrictions. On board data processing techniques such as automated mineral identification can ease bandwidth stress and increase scientific return. We have developed an automated support vector machine (SVM) detector operating in the VisNIR (300-2500 nm) spectral range trained to recognize the mineral jarosite (KFe$_{3}$(SO$_{4}$)$_{2}$(OH)$_{6}$). The detector input includes spectral wavelength intervals covering the primary jarositic spectral features at 620, 900 and 2280 nm and avoiding noisy features caused by atmospheric water vapor at 1400 and 1900 nm. The detector is trained on spectral library data (USGS speclib04) of 4 jarosite varieties and 85 samples of 21 non-jarosite minerals appropriate to Mars. To improve the training set, pure spectra were augmented with binary, tertiary and quaternary linear mixtures of spectra of the two (jarosite and non-jarosite) mineral groups. SVMs map training data into a high-dimensional kernel space and then fit a hyperplane that best separates the two classes of data. Initial results using spectra of pure (museum-quality) mineral samples taken in the laboratory include the correct identification of 16 jarosite spectra out of 209 diverse total spectra with no false negatives and one false positive. Results from laboratory spectra collected from field samples with mixed sulfate and phyllosilicate mineralogies include the correct detection of one jarosite and correct rejection of 13 clearly non-jarositic samples. Three of eight remaining samples were also detected as jarosite though the accuracy of these results will not be clear until the samples can be analyzed chemically. Future work will include the creation of detectors for other sulfate salts such as alunite (KAl$_{3}$(SO$_{4}$)$_{2}$(OH)$_{6}$) and other related minerals.
DE: 5464 Remote sensing
DE: 6094 Instruments and techniques
DE: 6225 Mars
SC: Planetary Sciences [P]
MN: 2004 AGU Fall Meeting