Performance of the Segment Anything Model in Various RFI/Events Detection in Radio Astronomy

Yanbin Yang,Feiyu Zhao,Ruxi Liang,Quan Guo,Junhua Gu,Yan Huang,Yun Yu

2024-10-30

Abstract:The emerging era of big data in radio astronomy demands more efficient and higher-quality processing of observational data. While deep learning methods have been applied to tasks such as automatic radio frequency interference (RFI) detection, these methods often face limitations, including dependence on training data and poor generalization, which are also common issues in other deep learning applications within astronomy. In this study, we investigate the use of the open-source image recognition and segmentation model, Segment Anything Model (SAM), and its optimized version, HQ-SAM, due to their impressive generalization capabilities. We evaluate these models across various tasks, including RFI detection and solar radio burst (SRB) identification. For RFI detection, HQ-SAM (SAM) shows performance that is comparable to or even superior to the SumThreshold method, especially with large-area broadband RFI data. In the search for SRBs, HQ-SAM demonstrates strong recognition abilities for Type II and Type III bursts. Overall, with its impressive generalization capability, SAM (HQ-SAM) can be a promising candidate for further optimization and application in RFI and event detection tasks in radio astronomy.

Instrumentation and Methods for Astrophysics

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to solve the problems of automatic detection and identification of radio - frequency interference (RFI) and events (such as solar radio bursts, SRB) in radio astronomy. With the sharp increase in the amount of data generated by radio telescopes, especially the construction of large radio telescopes such as the Square Kilometre Array (SKA), how to efficiently process this data and ensure its high quality has become a key challenge.

Specifically, the paper focuses on the following aspects:

1. **Reducing human intervention**: Traditional methods such as the SumThreshold method are effective, but in practical applications, they require manual parameter adjustment, which increases the human burden.
2. **Improving detection accuracy and efficiency**: Existing deep - learning methods rely on training data when dealing with RFI and have poor generalization ability, resulting in poor detection effects on unknown RFI.
3. **Dealing with complex and diverse RFI**: In reality, RFI sources are wide - ranging and complex, including broadcasting, digital television, satellites, meteor trails, lightning, aircraft communications, etc., which place higher requirements on the robustness and adaptability of the model.
4. **Developing highly general - purpose methods**: Existing deep - learning models are usually designed for specific tasks or telescopes and lack sufficient generalization ability, making it difficult to be directly applied to other fields.

To solve the above problems, the paper introduces the open - source image recognition and segmentation model - Segment Anything Model (SAM) and its optimized version HQ - SAM. These models are selected because of their strong generalization ability and zero - shot learning ability, and can perform well in different types of RFI detection tasks. Especially when dealing with large - area broadband RFI data, their performance is even better than that of the traditional SumThreshold method.

In addition, the paper also explores the performance of HQ - SAM in identifying different types of solar radio bursts (Type II and Type III eruptions), demonstrating its potential application value in the field of radio astronomy.

In short, this research hopes to find a more general, efficient and accurate RFI and event detection solution with minimal fine - tuning and structural adjustment by using the HQ - SAM (SAM) model, so as to meet the growing data processing needs in radio astronomy.