[Previous] | [Session 8] | [Next]
S. J. Bus, R. P. Binzel (MIT)
The second phase of the Small Main-belt Asteroid Spectroscopic Survey (SMASSII) yielded reflectance spectra for 1447 asteroids, covering the wavelength interval from 0.44 to 0.92 \mu m. Based on these data, a revised asteroid taxonomy was developed that takes advantage of the increased information contained in CCD spectra. Using the Tholen taxonomy as its foundation, the SMASSII taxonomy contains 26 classes that are defined solely on spectral absorption features (Bus, Ph.D. thesis, 1999).
Because the SMASSII taxonomy was defined using a combination of analytical techniques, as well as human judgement (through visual inspection of the data), application of this taxonomy to newly observed asteroids can be somewhat cumbersome. To simplify the procedure, a neural network has been designed and trained to automatically provide classifications for new asteroids. This network not only identifies the most probable taxonomic class for an object, but also proposes other possible classifications, along with confidence levels, when the spectrum lies close to one or more of the boundaries separating the classes. It is anticipated that access to this network and supporting algorithms will be made available via the world-wide web, so that any asteroid spectrum, covering the same wavelength interval as the SMASSII data, could be readily classified within this system. The design of this neural network will be described, along with discussions about its reliability and limitations.
This work was supported by NASA.