DRAFTS: A Deep Learning-Based Radio Fast Transient Search Pipeline

Yong-Kun Zhang,Di Li,Yi Feng,Chao-Wei Tsai,Pei Wang,Chen-Hui Niu,Hua-Xi Chen,Yu-Hao Zhu

2024-10-04

Abstract:The detection of fast radio bursts (FRBs) in radio astronomy is a complex task due to the challenges posed by radio frequency interference (RFI) and signal dispersion in the interstellar medium. Traditional search algorithms are often inefficient, time-consuming, and generate a high number of false positives. In this paper, we present DRAFTS, a deep learning-based radio fast transient search pipeline. DRAFTS integrates object detection and binary classification techniques to accurately identify FRBs in radio data. We developed a large, real-world dataset of FRBs for training deep learning models. The search test on FAST real observation data demonstrates that DRAFTS performs exceptionally in terms of accuracy, completeness, and search speed. In the re-search of FRB 20190520B observation data, DRAFTS detected more than three times the number of bursts compared to Heimdall, highlighting the potential for future FRB detection and analysis.

Instrumentation and Methods for Astrophysics,High Energy Astrophysical Phenomena

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the complex challenges in the detection of Fast Radio Bursts (FRBs). Specifically, the detection of FRBs is a complex task in radio astronomy, mainly due to the following two problems:

1. **Radio - Frequency Interference (RFI)**: Radio data in reality are often subject to radio - frequency interference from various man - made and natural sources, such as mobile phones, power lines, artificial satellites and lightning, etc. These interference signals are usually several orders of magnitude higher in intensity than astronomical signals, causing FRB signals to be masked.

2. **Signal Dispersion in the Interstellar Medium**: FRB signals will be dispersed due to the influence of the interstellar medium during propagation, stretching the originally only - a - few - milliseconds - long signal into data lasting several seconds, making it difficult to detect.

Traditional FRB search algorithms are inefficient, time - consuming, and will generate a large number of false positives. Therefore, the paper proposes a deep - learning - based radio fast transient source rapid search pipeline DRAFTS (Deep learning - based Radio Fast Transient Search Pipeline), aiming to identify FRB signals more efficiently and accurately.

### Main Features of DRAFTS

- **Multi - stage Pipeline**: DRAFTS integrates object detection and binary classification techniques and can effectively identify FRBs.
- **Large - scale Real - world Data Set**: Use the real FRB data observed by the FAST telescope to train the deep - learning model.
- **Efficient Processing**: Significantly improve the data processing speed through GPU acceleration (such as Numba.cuda).
- **Reduce False Positives**: Ensure that the same signal is not repeatedly detected through object detection, and perform secondary verification through the classification model, reducing the false positive rate.

### Specific Problems Solved

1. **Improve Detection Efficiency**: Compared with traditional methods, DRAFTS can complete data processing and FRB detection in a shorter time.
2. **Reduce False Positives**: Through effective training and optimization of the deep - learning model, the number of false positives is greatly reduced.
3. **Enhance Weak - signal Detection Ability**: DRAFTS can accurately identify weak FRB signals from a strong RFI background, solving the problem that traditional methods are difficult to detect weak signals.

In conclusion, DRAFTS significantly improves the accuracy and efficiency of FRB detection by combining deep - learning techniques and large - scale real - world data sets, providing strong support for future FRB research.