Galaxy Classification without Feature Extraction
Kai Lars Polsterer (Ruhr-Universität Bochum), Fabian Gieseke (Carl von Ossietzky Universität Oldenburg), Oliver Kramer (Carl von Ossietzky Universität Oldenburg)
The automatic classification of galaxies according to the different Hubble types is a widely studied problem in the field of astronomy. The complexity of this task lead to projects like Galaxy Zoo which try to obtain labeled data based on visual inspection by humans. Many automatic classification frameworks are based on artificial neural networks in combination with a feature extraction step in the preprocessing phase. These approaches rely on labeled catalogs for training the models. The small size of the typically used training sets, limits the generalization performance of the resulting models. In this work, we present a straightforward application of support vector machines for this type of classification tasks. The conducted experiments indicate that using a sufficient number of labeled objects provided by the EFIGI catalog leads to high-quality models. In contrast to standard approaches no additional feature extraction is required.
Paper ID: P119