Radio Frequency Interference Identification Model Of Radio Data Based On GAN

Radio Frequency Interference(RFI)inevitably affects radio astronomical observations,creating noise in the observations and affecting subsequent astronomical science research,so RFI needs to be identified and removed.The identification of RFI from radio observation data faces two main challenges: on the one hand,the increase of data complexity and the emergence of new RFI sources due to human activities require highly intelligent RFI identification methods;On the other hand,the increase of data scale and the increase of radio observation equipment capacity make the average daily observation data grow in terabytes,which requires a dedicated framework system to support large-scale data processing.To address the above challenges,this thesis proposes a deep learning-based RFI identification model,designs and implements an RFI data processing system for distributed heterogeneous computing environments.To address the problem of intelligence and adaptiveness of RFI recognition methods,this paper proposes an RFI recognition model based on generative adversarial networks(RFI-GAN),which transforms the RFI detection problem into an image feature recognition problem,and realizes high-precision RFI recognition.This paper designs and implements a multi-module,low-coupling data processing system,which can automatically schedule the RFI recognition processing tasks in the case of large-scale data,while ensuring that each processing task can be processed in the shortest possible time and obtain the shortest possible task completion time.The experimental results show that,compared with existing deep learning-based RFI recognition methods,the proposed RFI-GAN model has the best recognition effect,and the recognition Precision,Recall,and F1-Score of RFI all reach over 99%;The implemented distributed data processing system can effectively complete the assignment and execution of RFI identification tasks,and the proposed scheduling algorithm can achieve optimal scheduling of tasks and minimum task completion time for different task sizes and working nodes with different computational capabilities.