| |
 |
| Paper: |
U-NetIM: An Improved U-Net for Automatic Recognition of RFIs |
| Volume: |
523, Astronomical Data Analysis Software and Systems XXVIII |
| Page: |
123 |
| Authors: |
Long, M.; Yang, Z.; Xiao, J.; Yu, C.; Zhang, B. |
| Abstract: |
Radio frequency interference (RFI) mitigation is a key phase in data processing pipeline of radio telescopes. Classical RFI mitigation methods depending on the RFI physical characteristics can often fail to recognize some complicated patterns or result in misrecognition. We developed a novel approach of RFI recognition and automatic flagging using an improved convolution neural network. The improved U-Net model (U-NetIM) is constructed based on U-net with much deeper network structure for more complicated patterns and more components to reduce recognition error caused by over-fitting. The experiments show that the U-NetIM has better performance on both precision and recall rate than SumThreshold the most widely used classical method , U-Net, a traditional deep learning model and KNN,a typical machine learning model. |
|
|
|
|
|
