Searching for PNe in halpha surveys with AI methods

Abstract

Planetary nebulae (PNe) are the most complex and beautiful objects in modern astrophysics, which are essential to studying the late-stage evolution of low- to intermediate-mass stars and the chemical enrichment of the Galaxy. However, the known ∼ 3800 Galactic PNe are incomplete to uncover the complicated process of late-stage stellar evolution, which shows various observational characteristics such as morphologies and spectral identities. This thesis aimed to hunt for resolved PNe towards the Galactic center, which is not compact star-like nebulae obtainable by previous photometrical studies, but small and faint ones hidden in the dense star fields, also difficult for human eye inspection. Therefore, this thesis took advantage of the novel computer vision algorithm Swin-transformer’s efficiency and flexibility on multi-scale objects, trained the network with the PNe catalog provided by the HASH database, which was accumulated by hunting previous Hα survey, and merely thousands of IPHAS Hα images of the northern Galactic plane, and resulted in the precision rate of 96.5% and the recall rate of 97.8%. Then, this thesis applied the network to VPHAS+ high-resolution Hα imaging of the southern Galactic plane, a newly released survey without completed human eye inspection. After catalog comparison and eye screening, this thesis obtained 815 high-quality PN candidates, did follow-up spectral observations to 31 further selected top-quality samples on SAAO 1.9 m telescope, confirmed 70.97% of them are true, likely, and possible PNe by spectroscopy, and made further scientific analysis. This thesis capitalized on the newest Hα survey and AI techniques, which further completed the hunting of PNe, and provided more samples to promote progress in related fields. As an attempt at astronomical object detection with AI methods, this thesis explored the techniques that are applicable to multi-band and multi-modal astronomical data, which can contribute to the fusion of multi-band astronomy and the integration of the observation, data reduction, object detection, and classification of astronomical data.