WALLABY Pilot Survey: HI source-finding with a machine learning framework

The WALLABY HI survey with ASKAP generates vast data volumes, requiring reliable automation for source detection.We introduce a deep learning framework that improves automated identification of low SNR HI sources by filtering false positives in SoFiA candidate catalogues.Using 3D CNNs, our method effectively recognizes patterns in three-dimensional space, outperforming traditional linear approaches in accuracy and reducing false detections.Tests with mock galaxies in real ASKAP data show near-100% completeness and reliability at SNR ~3-5.Scaling this tool will enhance the scientific impact of future widefield HI surveys.