一种基于深度卷积神经网络的电磁干扰识别与抑制方法

A Method of Electromagnetic Interference Identification and Suppression based on Deep Convolutional Neural Network

随着无线电通信技术的发展,人工无线电对气象雷达的电磁干扰(EMI)明显增加,对雷达数据的质量产生严重影响。目前,关于检测和抑制电磁干扰的研究大多是基于雷达的初级产品。从雷达接收机前端的I/Q数据出发,提出使用深度卷积神经网络来识别和抑制电磁干扰的方法。设计一种残差结构的全卷积网络,并且选择UNet和DeepLab V3+共同进行识别效果的对比,在识别之后使用线性插值方法对电磁干扰进行抑制。结果显示,3种模型都能有效地识别电磁干扰,并且在识别的准确率上和召回率上各有优劣。在对识别结果进行抑制后,使得雷达数据质量都得到明显的提高。

In recent years,the development of radio communication technology has significantly increased Electromagnetis interference(EMI)from artificial radio,which has a detrimental impact on the quality of weather radar data.Current research on detecting and suppressing EMI mostly focuses on the primary radar products.This paper proposes a method to identify and suppress EMI using deep convolutional neural networks applied to the I/Q data of the front end of radar receivers.Specifically,this paper designed a fully convolutional network with residual structure.As well as UNet and DeepLab V3+were selected to compare the identification effect together,following which a linear interpolation method is used to suppress the electromagnetic interference.The results show that all three models can effectively identify EMI,and each has its advantages and disadvantages in the accuracy and recall rate.Suppressing EMI based on the identification significantly improves the quality of radar data.