Segmentation of the Galactic ISM Filaments using Deep Learning and Hi-GAL Catalogue

Understanding star formation is key to understand galaxy evolution. Observations collected in the Herschel Infrared Galactic plane Survey (Hi-GAL) revealed the ubiquitous presence of filamentary structures in the Galactic Plane. Filaments host star formation. Therefore, it is very interesting to study them and analyze the environment effect on their life cycle. A first step in understanding filaments consists in detecting them in the Galactic Plane. Detection algorithms based on standard image processing techniques present several limitations. Taking into consideration the drastic progress in Artificial Intelligence (AI), along with the data abundance about filaments, we propose to explore filament segmentation using Deep Learning (DL) framework. In this paper, we use state-of-the-art image segmentation in DL, U-Net, to segment filaments in all the Galactic Plane, in H2 column density images of the Galactic Plane obtained with Herschel Hi-GAL data. The used ground truth consists in the Hi-GAL filament catalogue provided by Schisano et al. Obtained results reveal more structures than provided in the catalogue, which makes AI-based methods very promising for such application.