Radio frequency interference detection based on the AC-UNet model
Ruiqing Yan, Cong Dai, Wei Liu, Jixia Li, Siying Chen, Xianchuan Yu, Shifan Zuo, Xuelei Chen
Abstract
Abstract Radio frequency interference (RFI) is a serious issue in radio astronomy. This paper proposes a U-Net network model with atrous convolution to detect RFI. Using the ability of convolutional neural networks to extract image features of RFI, and learning RFI distribution patterns, the detection model of the RFI is established. We use observational data containing real RFIs obtained by the Tianlai telescope to train the model so that the model can detect RFI. Calculate the probability of a data point being RFI pixel by pixel, and set a threshold. At the same time the dropout layer was added to avoid overfitting problems. If the predicted probability of a data point exceeds the threshold, it is considered that there is RFI, and if the predicted probability of a data point does not exceed the threshold, then it is considered that there is no RFI, so that the part of the image with RFI is flagged. Experimental results show that this approach can achieve satisfactory accuracy in the detection of radio observation images with a small amount of RFI.