| Name: Shivam Kumaran |
| Affiliation: Space Applications Centre, ISRO, Ahmedabad |
| Conference ID: ASI2026_219 |
| Title: Sutra: A ML based framework for Interstellar medium filament identification and beam-level characterisation |
| Abstract Type: Poster |
| Abstract Category: Facilities, Technologies and Data science |
| Author(s) and Co-Author(s) with Affiliation: Shivam Kumaran(Sapce Applications Centre, ISRO, Ahmedabad - 380015, India), Vipin Kumar(Sapce Applications Centre, ISRO, Ahmedabad - 380015, India), Ushasi Bhowmick(Sapce Applications Centre, ISRO, Ahmedabad - 380015, India), Manish Chauhan(Sapce Applications Centre, ISRO, Ahmedabad - 380015, India), Munn Vinayak Shukla(Sapce Applications Centre, ISRO, Ahmedabad - 380015, India), Mehul R Pandya(Sapce Applications Centre, ISRO, Ahmedabad - 380015, India) |
| Abstract: Filamentary structures in the interstellar medium (ISM) are essential to our understanding of how molecular clouds evolve and form stars. However, current methods for filament identification often require intensive manual parameter tuning or are computationally prohibitive for large-scale surveys. We introduce Sutra, a unified machine learning framework designed to automate the detection and physical characterization of these structures.
Sutra utilizes a U-Net convolution neural network architecture to provide a "parameter-free" identification process. The baseline model is trained on the union of outputs from existing tools (DisPerSE and GETSF), allowing it to detect a broader population of filaments, including faint, diffused structures often missed by traditional algorithms. The modular nature of Sutra allows us to incorporate other identification models, tuned across surveys such as HGBS, Hi-GAL, ATLASGAL as well as Position-Position-Velocity (PPV) data cubes.
From the identified skeleton map, the framework allows for both region wide filament properties characterization as well as detailed analysis of individual filament. To do this, radial profiles are extracted perpendicular to the local filament axis and then grouped into segments matching the given map’s beam resolution. This beam-level approach is crucial for studying local variations in linear mass density, local contrast, width, and Plummer p-index along the filament's length, revealing intricate physical changes that occur as filament approaches hub systems or undergo fragmentation.
We will demonstrate the application of Sutra using Herschel and APEX Column density maps, as well as 3D PPV maps. We will showcase Sutra’s ability to rapidly measure physical properties and provide a streamlined resource for research related to star formation. |
