Autonomous, rapid classifiers for hyperspectral imagers

Details Autonomous, rapid classifiers for hyperspectral imagers Autonomous, rapid classifiers for hyperspectral imagers

May 2006

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2006agusmin43e..05g&link_type=abstract

American Geophysical Union, Fall Meeting 2007, abstract #IN43E-05

Mathematics

Logic

6225 Mars, 6297 Instruments And Techniques

Scientific paper

Hyperspectral systems collect huge volumes of multidimensional data that require time consuming, expert analysis. The data analysis costs of global datasets restrict rapid classification to only a subset of an entire mission dataset, reducing mission science return. Data downlink restrictions from planetary missions also highlight the need for robust mineral detection algorithms. For example, both OMEGA and CRISM will map only approximately 5% of the Mars surface at full spatial and spectral resolution. While some targets are preselected for full resolution study, other high priority targets on Mars will be selected in response to observations made by the instruments in a multispectral survey mode. The challenge is to create mineral detection algorithms that can be utilized to analyze any and all image cubes (x, y, λ) for a selected system to help ensure that priority targets are not overlooked in these datasets. This goal is critical both for onboard, real time processing to direct target acquisition and for the mining of returned data. While an ultimate goal would be to accurately classify the composition of every pixel on a planet's surface, this is made difficult by the fact that most pixels are complex mixtures of n materials, which may or may not be represented in library (training) data. We instead focus on the identification of specific important mineral compositions within pixels in the data. For Mars, high priority targets include minerals associated with the presence of water. We have developed highly accurate artificial neural network (ANN) and Support Vector Machine (SVM) based detectors capable of identifying calcite (CaCO3) and jarosite (KFe3(SO4)2(OH)6) in the visible/NIR (350 to 2500 nm) spectra of both laboratory specimens and rocks in Mars analogue field environments. The detectors are trained using a generative model to create 1000s of linear mixtures of library end-member spectra in geologically realistic percentages. Here we will discuss preliminary results of the application of these classifiers to hyperspectral data sets. Since each mineral detector is focused on only a single mineral or mineral subclass, entire hyperspectral data cubes can be rapidly analyzed. This makes such detectors ideal for use in exploratory data analysis, where rapid feedback enhances discovery potential.

Affiliated with

Also associated with

No associations

Rating

Autonomous, rapid classifiers for hyperspectral imagers does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Autonomous, rapid classifiers for hyperspectral imagers, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Autonomous, rapid classifiers for hyperspectral imagers will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-1025447