The performances based on different input patterns were compared. Moreover, feature selection and feature weighting based on random forest were investigated. They then studied the discrimination of quasars from stars and the classification of quasars, stars, and galaxies with the sample from optical and radio bands and with that from optical and X-ray bands. The data was adopted from different databases, including SDSS data release five, USNO, FIRST, and ROSAT. introduced a decision tree method called random forest for multiwavelength data classification. Diagram of the neural star/galaxy classifier. If we want to improve the accuracy of automatic classification, it is necessary to increase the number of X-ray emitting stars because the majority of X-ray emitting sources are quasars and galaxies.įig. In other words, it is easy to separate quasars and galaxies, but it is difficult to discriminate X-ray active stars from quasars and galaxies. The result showed that the classification accuracy of quasars and galaxies adds up to more than 93% while that of X-ray emitting stars only amounts to 45.3%. Based on the identified samples with SDSS spectral classification, a random forest algorithm for automatic classification was performed. X-ray active stars have a similar distribution to most stars in the g–r versus r–i diagram. The earlier the type of star, the stronger its X-ray emitting. However, X-ray active stars have certain distributing rules according to spectral types. Quasars and galaxies occupy different zones while stars scatter in them. obtained one of the largest X-ray/optical catalogs and explored the distribution of various classes of X-ray emitters in the multidimensional photometric parameter space. The structure of a hierarchical NN.Ĭross-correlating the XMM-Newton 2XMMi-DR3 catalog with the SDSS data release, Zhang et al. The comparison of the NEXT performance with that of the best detection and classification package known to the authors (SEXTRACTOR) shows that NEXT is at least as effective as the best traditional packages.įig. In order to optimize the performance of the system, we implemented and tested several different NN models. We then used these selected features to perform the classification task. 14.1 (a three-layer fully connected neural network), to select the most significant features among the large number of measured ones. In order to obtain an objective and reliable classification, instead of using an arbitrarily defined set of features, the authors used an NN, shown in Fig. In practice, each object is classified depending on its membership of the regions mapping the input feature space in the training set. The authors therefore adopted a supervised NN (i.e., an NN that first finds the rules to classify objects from examples and then applies them to the whole dataset). Therefore, the parameters characterizing the two classes do not lie in disconnected subspaces, thus preventing the use of unsupervised methods. At magnitudes fainter than the completeness limit, stars are usually almost indistinguishable from galaxies. The first part of the NEXT procedure consists of an optimal compression of the redundant information contained in the pixels via a mapping from pixel intensities to a subspace individualized through PCA. In the extraction step, assuming that the images are well sampled, NEXT requires only the simplest a priori definition of “what an object is” (i.e., it keeps all structures composed of more than one pixel) and performs the detection via an unsupervised NN, approaching detection as a clustering problem that has been thoroughly studied in the artificial intelligence literature.

They are then classified as stars or galaxies through diagnostic diagrams having variables chosen according to the astronomer's taste and experience. Traditionally in astronomical images, objects are first distinguished from the noisy background by searching for sets of connected pixels having brightness above a given threshold. presented NEXT (Neural Extractor), a new neural network (NN)-based package capable of detecting objects and performing both deblending and starting galaxy classification in an automatic way. And the more data we provide, the more accurate this is. In classification and detection tasks, supervised learning can utilize data with labels to construct models.

Blazar vs quasar