Accelerating FRB Search: Dataset and Methods

TL;DR

This paper introduces the FAST-FREX dataset and a novel machine learning algorithm, RaSPDAM, significantly improving the efficiency and accuracy of Fast Radio Burst detection using radio telescope data.

Contribution

The paper presents a new dataset and a machine learning method that outperforms traditional algorithms in FRB detection, advancing the field's capabilities.

Findings

RaSPDAMv2 achieves 97% precision and 83% recall.

The new dataset includes 600 positive and 1000 negative samples.

RaSPDAM outperforms PRESTO and Heimdall in accuracy and efficiency.

Abstract

Fast Radio Burst (FRB) is an extremely energetic cosmic phenomenon of short duration. Discovered only recently and with its origin still unknown, FRBs have already started to play a significant role in studying the distribution and evolution of matter in the universe. FRBs can only be observed through radio telescopes, which produce petabytes of data, rendering the search for FRB a challenging task. Traditional techniques are computationally expensive, time-consuming, and generally biased against weak signals. Various machine learning algorithms have been developed and employed, all of which require substantial datasets. We here introduce the FAST dataset for Fast Radio bursts EXploration (FAST-FREX), built upon the observations obtained by the Five-hundred-meter Aperture Spherical radio Telescope (FAST). Our dataset comprises 600 positive samples of observed FRB signals from three sources and 1000 negative samples of noise and Radio Frequency Interference (RFI). Furthermore, we provide a machine learning algorithm, Radio Single-Pulse Detection Algorithm Based on Visual Morphological Features (RaSPDAM), with significant improvements in efficiency and accuracy for FRB search. We also employed the benchmark comparison between conventional single-pulse search softwares, namely PRESTO and Heimdall, and RaSPDAM. RaSPDAMv2 achieves an average precision of 97% and an average recall of 83%, with notable enhancements in computational performance. Future machine learning algorithms can use this as a reference point to measure their performance and help the potential improvements. By enabling more accurate and efficient detection of transient radio events, our work facilitates the FRB and pulsars search pipeline, enhances the potential for discovering new astrophysical phenomena.