Automatic Classification of Radio Sources in the Galactic Plane Using B-CNN

The ever-increasing number of sky surveys calls for the need for more efficient data processing alternatives. CNN has been widely used in computer vision and is currently being incorporated in classifying astronomical objects. A Branch Convolution Neural Network (B-CNN) is a hierarchical-based Convolution Neural Network (CNN) that outputs multiple predictions of coarse to fine levels. This thesis involved developing a B-CNN-based classifier for, automatic classification of objects discovered in the Co-Ordinated Radio ’N’ Infrared Survey for High-mass star formation (CORNISH) south survey. First, a 2-level model was developed and trained on similar CORNISH north sources, with additional multi-wavelength data from the IR and the sub-millimetre surveys. From the catalogue, sources used were the ultra-compact HII (UCHII) regions, HII regions, planetary nebulae (PN), radio stars, and extra-galactic background sources. The developed classifier was then tested on a sample of already classified CORNISH south sources with an accuracy of 67% for the coarse level and 33% for the fine level.