[Previous] | [Session 15] | [Next]
W.B. Weaver (MIRA)
An artificial neural network (ANN) technique has been developed to perform two-dimensional classification of the components of binary stars of any temperature or luminosity classifications. Using 15 Angstrom-resolution spectra, a single ANN can classify the unresolved components with an average accuracy of 2.5 subclasses in temperature and about 0.45 classes in luminostiy for up to 3 magnitudes difference in luminosity. The use of two ANNs, the first providing coarse classification while the second provides specialist classification, reduces the mean absolute errors to about 0.5 subclasses in temperature and 0.33 classes in luminosity.
The system operates with no human intervention except initial wavelength registration and can classify about 20 binaries per second on a Pentium-class computer.
This research was supported by the Friends of MIRA.
The author(s) of this abstract have provided an email address for comments about the abstract: bw@mira.org