Research On The Identification Of Radio Frequency Interference Based On Machine Learning

Square Kilometre Array(SKA)is the largest telescope under construction.This great project requires sufficient simulation in the early stage to guide its subsequent construction.Flagging of Radio Frequency Interference(RFI)is a significant challenge for radio astronomy.We use Radio Astronomy Simulation Calibration and Imaging Library(RASCIL)of SKA.We also reference the RFI simulation method of the Hydrohydrogen Epoch of Reionization Array(HERA)to create various RFI.Machine Learning is applied to identify RFI in our work.In this paper,the first chapter introduces the background of radio data storage,we also discuss the current methods of flagging RFI.In the second chapter,a method of generating a complete MS file by using Python-casacore package was proposed to meet the practical application requirements.In the third chapter,the observation principle of the radio interferometer array is introduced in detail and the method of simulating the observation data by using RASCIL is given.In the fourth chapter,Le-Net,a deep convolutional neural network architecture,is proposed for the flagging of RFI.We compare the robustness between our Le-Net model and AOflagger,which is a flagger framework that implements several methods to deal with RFI.Our Le-Net model achieves a Accuracy of98.8%,Recall of 89.3% and Precision of 99% as applied to our simulation data.In the fifth chapter,we discuss the identification of weak RFI.Our results suggest the advantage of deep learning applied in radio astronomy.It likely play a key role in the future of radio astronomy data.