Abstract:Abstract In the realm of radio astronomy, the detection of Radio Frequency Interference (RFI) stands as a pivotal pursuit. This study performs a novel comparative analysis of deep learning methodologies, and introduces a novel transfer learning method, called fine-tuning. We compare various aspects and factors relating to this problem, including supervised Fully Convolutional Network (FCN) architectures used within the literature, loss functions, regularization techniques, and training methodologies, to establish the most effective strategies for RFI detection. Moreover, the relationship between parameters, FLOPS, and inference times are examined. Fine-tuning involves pre-training models with low-quality AOFlagger reference outputs, a very popular and accessible RFI flagging software package, and thereafter re-training the models with high-quality reference outputs. We utilize two datasets: real observations from LOFAR and simulated data from HERA. The Mean Squared Error (MSE) loss function emerges as a robust performer if a high recall is desired. In contrast, the Binary Cross-Entropy (BCE) loss function excels in generalization but falls short in classification performance. The Dice loss function emerges as the top performer, maximizing the F1 score and thereby serving as the choice for our further investigations. Notably, we highlight the important role of data quality and model capacity. In particular, we find that low-capacity models exhibit resilience when trained with low-quality flags from AOFlagger, showcasing their ability to mitigate overfitting and overflagging tendencies. In contrast, high-capacity models excel when trained with high-quality flags. Fine-tuning proved to be an effective method to unlearn the overflagging tendencies of AOFlagger, whilst requiring very little data.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: in the field of radio astronomy, how to effectively detect radio - frequency interference (RFI). Specifically, by comparing different deep - learning methods, the researchers introduced a new transfer - learning method - fine - tuning, to improve the accuracy and robustness of RFI detection.
### Problem Background
Radio - frequency interference (RFI) means that electromagnetic signals interfere with the celestial - body signals of interest. Therefore, accurate detection of RFI is crucial for radio astronomy. RFI can come from multiple sources, such as satellites, radars, and telecommunication devices. These signals are much stronger than those from celestial bodies, and a few seconds of unprocessed RFI can seriously affect the observations. Although interference sources can be restricted by building radio - quiet zones or international regulations, it is impossible to completely avoid RFI, so effective detection and processing methods are required.
### Existing Challenges
1. **Diversity of RFI**: RFI can take on multiple forms, and each form has a different impact on scientific goals.
2. **Large amount of data**: The data transfer rate of modern radio telescopes is very high (300 GB/s to 19 TB/s), requiring automated processing.
3. **Limitations of existing methods**:
- Although statistical methods (such as CUMSUM, SVD, SumThreshold) are widely used, they still require human intervention in practical applications.
- Although deep - learning methods can improve efficiency, they have problems such as over - fitting and insufficient ability to adapt to complex RFI patterns.
### Research Objectives
To address the above challenges, this research mainly focuses on the following aspects:
1. **Evaluating the performance of different loss functions**: including mean - squared error (MSE), binary cross - entropy (BCE), Dice loss, and Log - Cosh - Dice loss.
2. **Analyzing the impact of the quality of reference training data on model performance**: especially the impact of low - quality labels (such as AOFLAGGER output) and high - quality labels (such as expert annotations).
3. **Exploring training methods**: especially introducing a new transfer - learning method - fine - tuning, that is, first pre - training the model with low - quality labels, and then fine - tuning it with high - quality labels.
4. **Evaluating computational complexity**: including floating - point operations (FLOPS), the number of parameters, and execution time, etc.
### Data Sets
The research used two data sets:
- **LOFAR real - observation data**: Actual observation data from the LOFAR telescope.
- **HERA simulation data**: Simulation data generated using the hera_sim library, containing multiple RFI types and noise.
### Main Contributions
1. **Evaluating the classification performance and generalization ability of four loss functions**.
2. **Analyzing the impact of regularization techniques on classification performance and generalization ability**.
3. **Exploring the trade - off between classification performance and computational complexity**.
4. **Systematically evaluating the classification performance and generalization ability of different architectures and correlating them with model capacity**.
5. **Introducing the fine - tuning method** and demonstrating its effectiveness in reducing over - fitting and improving performance.
6. **Proposing two new CNN architectures**: ASPP - Unet and DSC - Mono - Resunet, providing valuable insights into computational complexity.
Through these efforts, the researchers hope to provide more effective methods for RFI detection and promote the application of deep learning in radio astronomy.