Application of Genetic Programming to Identifying Asteroid Surface Composition from Reflectance Spectra

K. A. Farry (NASA/JSC/NRC), F. Vilas (NASA/JSC), K. S. Jarvis (LMSMSS)

We are evaluating the usefulness of genetic programming in identifying the minerals shaping the spectra of light reflected from low-albedo asteroids. Genetic programming is an evolutionary programming method that has proven useful in complex classification problems in other fields such as biology (e.g., identifying hand motions from measured variations in muscle electric fields [Fernandez et al, Genetic Programming-96 Conference, Stanford University, July 28-31, 1996]).

Our first effort to test the basic feasibility of this method for mineral identification involves identifying three terrestrial minerals likely to be present in asteroid regoliths from laboratory-produced spectra which we have corrupted with simulated incomplete atmospheric corrections and instrumentation noise. This poster will cover progress in this phase of the feasibility study. Our next steps will be expanding the number of minerals identified and altering the laboratory spectra to the lower resolution and range-limited form of much of the asteroid reflectance data already collected (e.g., ECAS and 52-color survey). We will use this altered data set to evolve a mineralogical classifier which we will apply to actual asteroid data.