Extracting features by interpolating and down-sampling for Galaxy and QSO spectrum classification

For celestial spectra are vectors in a several-thousand-dimensional space with a mass of redundancy and usually contaminated with various noises, feature extraction is an essential procedure in automatic spectra processing. We investigated the feature extraction problem for Quasar and Galaxy spectra classification. The available methods in literature can be loosely classified into the following types: principal component analysis (PCA), wavelet transform, artificial neural networks (ANN), Rough Set and Bayesian mixed model. In this work, by analyzing the traditional feature extraction methods, we introduced a novel feature analysis framework STP (Space Transformation and Partition) and proposed a novel feature extraction method EFCD (Extracting features by curve-fitting and down-sampling). Researches also show that it is sufficient to extract features in some cases, not necessary to use the sophisticated methods, which is usually more complex in computation. The proposed EFCD method is evaluated by recognizing Galaxy and QSO spectra, which is disturbed by red shift and representative in automatic spectra classification research. The results of this work are helpful to gain novel insight into the traditional feature extraction methods and design more efficient spectrum classifying schemes.