基于时频谱图和CNN的雷达空中目标识别方法  

An Aerial Target Recognition Method for Radar Based on Time-Frequency Spectrogram and CNN

作　　者：潘美艳 蔡兴雨[1] 薛健 PAN Meiyan;CAI Xingyu;XUE Jian(Xi'an Electronic Engineering Research Institute,Xi'an 710100;Xi'an University of Posts and Telecommunications,Xi'an 710121)

机构地区：[1]西安电子工程研究所,西安710100 [2]西安邮电大学,西安710121

出　　处：《火控雷达技术》2023年第4期16-22,共7页Fire Control Radar Technology

基　　金：国家自然科学基金项目(62201455);陕西省教育厅科研计划项目(22JK0566);陕西省科学技术协会青年人才托举计划项目(20230112)。

摘　　要：传统雷达空中目标识别方法容易受到杂波和有效特征难挖掘等因素影响,其识别准确率难以满足实际雷达系统的目标识别需求。为了改善雷达对空中目标的识别性能,提出一种基于时频谱图和卷积神经网络(convolutional neural network,CNN)的空中目标识别方法。该方法首先提取雷达目标检测点迹所在距离单元的多脉冲回波数据,然后通过杂波抑制和基于CLEAN算法的雷达目标机身分量消除,得到以目标微多普勒分量为主的时域回波数据,接着通过短时傅里叶变换得到目标微多普勒时频谱图,最后利用时频谱图对设计的卷积神经网络进行训练和测试,实现喷气式飞机、直升机、螺旋桨飞机三类典型空中目标的识别。实测雷达数据测试结果表明,所提方法对三类飞机目标的识别准确率高于传统方法。所提方法避免了传统人工提取微动统计特征,消除了杂波与机身分量的影响,提高了雷达对空中目标识别的性能。Traditional radar aerial target recognition methods are easily affected by factors such as clutter and difficulty in mining effective features,and their recognition accuracy cannot meet the target recognition needs of actual radar systems.In order to improve the aerial target recognition performance of radar,an aerial target recognition method based on time-frequency spectrogram and convolutional neural network(CNN)is proposed.The proposed method first extracts the multi-pulse echo data of the range bin where the radar target detection is located,and then the time domain echo data dominated by the target micro-Doppler component is obtained through clutter suppression and elimination of the radar target airframe component based on the CLEAN algorithm.After that,the target micro-Doppler time-frequency spectrograms are obtained by using short-time Fourier transform.Finally,the time-frequency spectrograms are used to train and test the designed CNN to realize recognition of three typical types of aerial targets,namely jet aircrafts,helicopters,and propeller-driven aircrafts.Results of the test using radar measured data show that the proposed method has higher recognition accuracy for the three types of aircraft targets than the traditional methods.The proposed method avoids the manual extraction of micro-motion statistical features and eliminates the influence of clutter and airframe component,improving the aerial target recognition performance of radar.

关 键 词：卷积神经网络 空中目标识别 CLEAN算法 微多普勒 时频谱图 

分 类 号：TN95[电子电信—信号与信息处理]