Executive summary of the 17 page paper referenced by Robin:
Spectra of 100,000 stars was used to train a computer to recognise the spectral classes (OBAFGKM etc) using three different algorithms (PNN, SVN and K-means).
These algorithms were then used to identify a batch of 300,000 spectra.
It was necessary to use 700 data points on the spectrum between 3850 and 8900A.
These algorithms have been able to correctly classify approximately 80% of the data set.
It took between 23 and 29hr computing time to do the analysis.
|