Radio frequency interference detection using efficient multiscale convolutional attention UNet

Fei Gu,Longfei Hao,Bo Liang,Song Feng,Shoulin Wei,Wei Dai,Yonghua Xu,Zhixuan Li,Yihang Dao

DOI: https://doi.org/10.1093/mnras/stae868

IF: 4.8

2024-03-23

Monthly Notices of the Royal Astronomical Society

Abstract:ABSTRACT Studying the Universe through radio telescope observation is crucial. However, radio telescopes capture not only signals from the universe but also various interfering signals, known as radio frequency interference (RFI). The presence of RFI can significantly impact data analysis. Ensuring the accuracy, reliability, and scientific integrity of research findings by detecting and mitigating or eliminating RFI in observational data, presents a persistent challenge in radio astronomy. In this study, we proposed a novel deep learning model called EMSCA-UNet for RFI detection. The model employs multiscale convolutional operations to extract RFI features of various scale sizes. Additionally, an attention mechanism is utilized to assign different weights to the extracted RFI feature maps, enabling the model to focus on vital features for RFI detection. We evaluated the performance of the model using real data observed from the 40 m radio telescope at Yunnan Observatory. Furthermore, we compared our results to other models, including U-Net, RFI-Net, and R-Net, using four commonly employed evaluation metrics: precision, recall, F1 score, and IoU. The results demonstrate that our model outperforms the other models on all evaluation metrics, achieving an average improvement of approximately 5 per cent compared to U-Net. Our model not only enhances the accuracy and comprehensiveness of RFI detection but also provides more detailed edge detection while minimizing the loss of useful signals.

astronomy & astrophysics

What problem does this paper attempt to address?

The paper aims to address the issue of Radio Frequency Interference (RFI) detection in radio astronomical observations. Since radio telescopes capture not only signals from the cosmos but also various interference signals, i.e., RFI, these interferences can severely affect the accuracy, reliability, and completeness of data analysis and scientific research results. Therefore, accurately detecting and eliminating RFI in observational data is a continuous challenge in radio astronomy. To solve this problem, the research team proposed a new deep learning model—EMSCA-UNet. This model extracts RFI features of different scales through multi-scale convolution operations and uses an attention mechanism to assign different weights to the extracted RFI feature maps, enabling the model to focus on key features for RFI detection. Additionally, the study evaluated the model using actual observational data from the 40-meter radio telescope at Yunnan Astronomical Observatory and compared it with other models such as U-Net, RFI-Net, and R-Net on four commonly used evaluation metrics: accuracy, recall, F1 score, and IoU. Experimental results show that the EMSCA-UNet model outperforms other models on all evaluation metrics, with an average improvement of about 5% compared to U-Net. This indicates that the new model not only improves the accuracy and comprehensiveness of RFI detection but also provides more detailed edge detection while minimizing the loss of useful signals.