Dr Tim Galvin1, Dr Minh Huynh1
1CSIRO Astronomy and Space Science, Kensington, Australia
The classification of radio galaxies remains a challenge for radio astronomy. Large surveys such as Evolutionary Map of the Universe (EMU) on the Australian Square Kilometre Array Pathfinder (ASKAP) will detect 10s of millions of radio galaxies, increasing the number of known radio sources by an order of magnitude. The best current method for classifying radio sources as a single radio galaxy or multiple components of a single radio galaxy, and identifying optical/infrared host galaxies, is human eyeballing. This will not be feasible with the new EMU survey, so machine learning techniques are crucial for extracting the science from this radio survey. Parallelized rotation and flipping Invariant Kohonen maps (PINK, Polsterer et al.) is a self-organised map method of reducing the dimensionality of the data and reduce the number of images which will need to be classified by the human eye. We will present the initial results of using the PINK algorithm on radio-IR data cubes and compare the results to a classifier based on Faster Region-based Convolutional Neutral Network (Faster R-CNN).
Biography:
Minh Huynh graduated from ANU with a PhD in Astronomy and Astrophysics in 2005. She worked in NASA’s Spitzer Space Telescope and Planck Observatory teams at Caltech in Los Angeles, before moving back to Perth as Deputy International Project Scientist for the Square Kilometre Array at the University of Western Australia. The CSIRO is busy commissioning the Australian SKA Pathfinder (ASKAP), a next generation radio telescope. Now at the CSIRO in Perth, Minh continues astronomical research on radio galaxies while also preparing CSIRO’s ASKAP Science Data Archive for the big data deluge from ASKAP.