Source finding, parametrization and classification for the extragalactic Effelsberg-Bonn HI Survey

TL;DR

This paper presents an automated pipeline for source finding, parametrization, and classification in large-scale HI surveys, demonstrating comparable or superior accuracy to manual methods through simulations.

Contribution

It introduces a fully automated, wavelet-based source detection and neural network classification pipeline tailored for future large-scale HI surveys.

Findings

Automated pipeline achieves high accuracy in source parametrization.

Neural network classifier effectively distinguishes real sources from false positives.

Performance matches or exceeds manual methods used in HIPASS survey.

Abstract

Context. Source extraction for large-scale HI surveys currently involves large amounts of manual labor. For data volumes expected from future HI surveys with upcoming facilities, this approach is not feasible any longer. Aims. We describe the implementation of a fully automated source finding, parametrization, and classification pipeline for the Effelsberg-Bonn HI Survey (EBHIS). With future radio astronomical facilities in mind, we want to explore the feasibility of a completely automated approach to source extraction for large-scale HI surveys. Methods. Source finding is implemented using wavelet denoising methods, which previous studies show to be a powerful tool, especially in the presence of data defects. For parametrization, we automate baseline fitting, mask optimization, and other tasks based on well-established algorithms, currently used interactively. For the classification of candidates, we implement an artificial neural network which is trained on a candidate set comprised of false positives from real data and simulated sources. Using simulated data, we perform a thorough analysis of the algorithms implemented. Results. We compare the results from our simulations to the parametrization accuracy of the HI Parkes All-Sky Survey (HIPASS) survey. Even though HIPASS is more sensitive than EBHIS in its current state, the parametrization accuracy and classification reliability match or surpass the manual approach used for HIPASS data.