Detecting HI Galaxies with Deep Neural Networks in the Presence of Radio Frequency Interference

Abstract

In neutral hydrogen (HI) galaxy survey, a significant challenge is to identify and extract the HI galaxy signal from observational data contaminated by radio frequency interference (RFI). For a drift-scan survey, or more generally a survey of a spatially continuous region, in the time-ordered spectral data, the HI galaxies and RFI all appear as regions which extend an area in the time-frequency waterfall plot, so the extraction of the HI galaxies and RFI from such data can be regarded as an image segmentation problem, and machine learning methods can be applied to solve such problems. In this study, we develop a method to effectively detect and extract signals of HI galaxies based on a Mask R-CNN network combined with the PointRend method. By simulating FAST-observed galaxy signals and potential RFI impacts, we created a realistic data set for the training and testing of our neural network. We compared five different architectures and selected the best-performing one. This architecture successfully performs instance segmentation of HI galaxy signals in the RFI-contaminated time-ordered data (TOD), achieving a precision of 98.64% and a recall of 93.59%.