Radio Frequency Interference Detection Using Swin Transformer Embedding U2-Net
Radio frequency interference (RFI) is radio wave interference from natural sources or man-made models. In radio astronomy research, the signals of celestial objects captured by radio telescopes are extremely weak, and the presence of RFI can significantly mask or distort those signals, reducing the...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2025-01-01
|
| Series: | Advances in Astronomy |
| Online Access: | http://dx.doi.org/10.1155/aa/3232269 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Radio frequency interference (RFI) is radio wave interference from natural sources or man-made models. In radio astronomy research, the signals of celestial objects captured by radio telescopes are extremely weak, and the presence of RFI can significantly mask or distort those signals, reducing the accuracy of observational data and seriously affecting the reliability of scientific conclusions. Therefore, accurate RFI detection from radio astronomy data is of great importance. Currently, most RFI detection methods still rely on traditional methods, but due to the limitations of these methods, they are unable to accurately detect RFI in radio telescope observation data. To this end, we propose a novel deep learning–based RFI detection model named ST-U2Net, which combines the Swin Transformer and the Residual U-block (RSU) of U2-Net to form a dual-encoder architecture. The main encoder enhances the feature representation through the efficient multiscale attention (EMA) mechanism, reorganizes the channel information, captures pixel-level relationships, and improves the detection accuracy in complex backgrounds. The auxiliary encoder introduces a spatial interaction module (SIM) and a feature compression module (FCM) to enhance the feature representation and reduce the detail loss of narrowband RFI, respectively. In addition, the multilayer perceptron (MLP) in Swin Transformer is replaced by Kolmogorov–Arnold network (KAN) to enhance the modeling capability of narrowband RFI features. The relational aggregation module (RAM) fuses the characteristics of the two encoders to achieve a more accurate detection of RFI. In this study, the proposed ST-U2Net model is verified using actual observation data collected by the 40-m radio telescope at Yunnan Observatory. The experimental results show that compared to existing deep learning methods, the model proposed in this study achieves a significant improvement in the accuracy of detecting RFI, especially in the detection of narrowband RFI, which exhibits obvious advantages. |
|---|---|
| ISSN: | 1687-7977 |