Classifying Seyfert galaxies with deep learning

Seyfert galaxies have several subclasses according to observation features of their optical spectra. Traditional classification is usually visual inspection or using a quantity that is defined as a flux ratio between the Balmer line and forbidden line. However, visual inspection is time-consuming and a quantity derived from flux ratio does not reflect the information of emission line shape which is the original classification feature. One algorithm of deep learning is called Convolution Neural Network (CNN) and has shown several successful classification results. We use a CNN model to classify spectra of the Seyfert galaxies by building a 1-dimension CNN model as a classifier to distinguish Seyfert 1.9 galaxies from Seyfert 2 galaxies. We find our model can recognize Seyfert 1.9 and Seyfert 2 galaxies correctly with an accuracy of over 80%. Our model picks out an additional Seyfert 1.9 sample which was missed by visual inspection among Seyfert 2 samples in a short time. We also use the new Seyfert 1.9 sample to improve the performance of our model and obtain 91% precision of the Seyfert 1.9. This result indicates that our model can act as a classifier for picking out Seyfert 1.9 spectra among Seyfert 2 spectra in an efficient way for the future. The second part of the thesis is about the emission line of our Seyfert 1.9 galaxies. We decompose the H alpha emission line into narrow component and broad component by fitting 2 Gaussian components. We use the fitting results to derive line flux and line width to investigate the properties of emission lines of our Seyfert 1.9 galaxies. We find the 641 Seyfert 1.9 sample has an averagely luminous broad H alpha component than the 656 Seyfert 1.9 sample. This result indicates that our model can pick out sources with relatively weak broad H alpha component and these sources are sometimes missed by visual inspection. We investigate the black hole mass, host galaxy morphology, BPT diagram of the two Seyfert 1.9 samples. We find that the distributions of supermassive black hole (SMBH) mass of our Seyfert 1.9 galaxies are dominated by 10^7 solar mass. We find the two Seyfert 1.9 samples have slightly different black hole mass distributions but the t-test results indicate that the two distributions have similar mean values. From host galaxy results, we find the distribution of the host galaxy morphology of the Seyfert 1.9 galaxies is dominant by large bulge galaxy. We find the BPT distributions of our Seyfert 1.9 samples are similar to Seyfert 2 galaxies. This result indicates that Seyfert 1.9 and Seyfert 2 galaxies have similar ionization levels in their narrow-line region. Besides, we cross-search the Strip 82 survey with our Seyfert 1.9 sources for multiple spectra to investigate the possible variability. We only have 13 Seyfert 1.9 sources that have multiple spectra and find 13 Seyfert 1.9 sources show variation in continuum and broad H alpha component.