Asteroid taxonomy using artificial neural networks

Asteroids are classified into different taxonomic groups according to their spectral reflectance properties in the visual and near-infrared (Vis-NIR) wavelengths. The spectral properties of the asteroid surfaces can be related to the material of the surface. There are a few taxonomic systems for asteroids, the most recent being the so-called Bus-DeMeo taxonomy (B-DM, DeMeo et al., Icarus 202, 2009). Usually, the exact wavelengths used in the taxonomic system are tied up with the particular survey data that was used to create the taxonomy. With the B-DM taxonomy, it is the SMASSII survey extended to near-infrared with the NASA IRTF telescope observations, and the wavelengths are 0.45–2.45 µm.

The ESA space observatory Gaia will produce a significant number of low-resolution asteroid spectra over the 0.33–1.05 µm wavelength range in the Data Release 3 and the final data releases. There is an evident need for evaluating the surface properties of these asteroids using the Gaia data. For example, the list of taxonomic classifications for asteroids, maintained in the NASA Planetary Data System, has 2,600 asteroids with some taxonomic class (Neese, C., Ed., NASA Planetary Data System, 2010), but the Gaia data will eventually contain about 100,000 asteroid spectra.

There is a plan to provide a new asteroid taxonomy using the Gaia observations (Delbo et al., PSS 73, 2012). However, a link to existing taxonomic systems such as B-DM would be highly valuable. In this work, we study the possibility to use feed-forward artificial neural network for asteroid spectral classification. We show that the classification accuracy can remain on a reasonably good level even if the B-DM classification is done with the neural network that is trained to use only data having wavelengths in the 0.45–1.05 µm range, which is the overlapping region with the Gaia and the original B-DM systems. This tool can provide the B-DM taxonomic classification for all the asteroids with Gaia spectroscopy.

Acknowledgements: Research supported, in part, by the Academy of Finland (project 325805).