Stellar Spectral Classification with Machine Learning Classifier Systems and Chebyshev Coefficients using the Tonantzintla Schmidt Camera Plate Collection

We propose a method for automatic stellar spectral classification using a large collection of astronomical plate images improving the number of true positives for several spectral type classes in comparison with other approaches using the WND-CHARM feature extractor in order to study the most relevant parameters for this problem, concluding that Chebyshev Coefficients applied to images are the best alternative to describe up to seven different Stellar Spectral Types according to the Harvard Spectral Classification system taking account the main classes [O, B, A, F, G, K, M] sorted from the hottest to the coldest stellar object by their chemical composition. We present precision and accuracy rates reached by multiple machine learning approaches, obtaining applicable results for five and seven of the most important stellar spectral classes.