A Machine-learning Model to Separate Stars and Galaxies in iPTF Images
Abstract
The Intermediate Palomar Transient Factory (iPTF) is a dedicated time-domain survey optimized for the rapid characterization of fast transients. While significant efforts have been devoted to the development of software that quickly and reliably identifies new transients, there are currently no mechanisms to automatically classify these sources. The first component in deriving a classification is understanding whether or not the newly discovered transient is galactic or extragalactic in its origin. Here, we present our development of a new framework for classifying sources in iPTF reference images as either stars or galaxies. The framework utilizes the random forest algorithm and is trained with nearly 3 million sources that have Sloan Digital Sky Survey (SDSS) spectra. The final optimized model achieves a cross-validation accuracy of ~96%, which represents a significant improvement over the automated classification provided by the SExtractor algorithm. This accuracy, while slightly worse than that provided by the SDSS photometric classifier, can be extended over the entire iPTF footprint, which covers >5000 deg^2 that have not been imaged by SDSS. Associating transients with galactic or extragalactic origin is the first step in delivering automated classifications of newly discovered transients.
- Publication:
-
American Astronomical Society Meeting Abstracts #227
- Pub Date:
- January 2016
- Bibcode:
- 2016AAS...22734911M