基于一维距离像和微动特征的弹道导弹识别方法

A recognition method of ballistic missile based on multiple feature combination

弹道导弹目标识别是导弹防御系统的重要环节.用于弹道导弹识别的特征通常包括RCS特征、极化特征、雷达高分辨一维距离像(HRRP)、ISAR像、微动特征等.研究目标散射中心模型,给出了基于散射中心理论的HRRP特征.弹道导弹在释放诱饵过程中,弹头和诱饵在结构和质量上的差别会表现在微运动特征上,据此构建了弹头及诱饵的微运动模型,得到弹头进动模型及诱饵摆动模型,对所建模型回波进行短时傅里叶变换(STFT),提取时频脊得到微动特征.弹头及其诱饵有着丰富的结构信息,仅依靠单一特征识别弹头目标效果并不理想.提出综合HRRP特征与微动特征的多特征联合识别方法,对联合的多特征采用主成分分析(PCA)降维,采用支持向量机(SVM)作为分类算法,仿真结果表明,联合HRRP特征与微动特征并通过PCA降维后弹头识别率得到明显提升.

The purpose of this paper is to increase the recognition rate of proj ectile target.For identifying proj ectile target,it is necessary to extract target feature.Ballistic missile features commonly used in the scientific research work include radar cross section（RCS）feature,polarization feature,high resolution range profile（HRRP）,inverse synthetic aperture radar（ISAR）image,micro-motion feature etc.This paper mainly elaborates HRRP and micro-motion feature.In order to simplify the calculation of HRRP,we apply target scattering centers model.Ballistic missile release the decoy to avoid interception during the flight,causing warhead’s motion behaves as procession while decoy’s motion behaves as swaying.This study analyses how warhead and decoy moves when ballistic missile release the decoy and builds models for them.We process micro-motion model with short-time Fourier transform and extract the time-frequency ridge as micro-motion feature.Warhead and decoy are abundant in structure information so it is not proper to identify proj ectile target merely rely on one single feature.This paper puts forward a superior recognition method based on multiple feature combination,combining HRRP feature with micro-motion feature.We＆amp;nbsp;use principal component analysis（PCA）to reduce the dimensionality and choose support vector machine（SVM）as classification algorithm.We draw HRRP figure and multiple feature spectrum figure for the warhead and the decoy through the simulation experiment.In order to simulate the target recognition process accurately,this paper change the number of test set and create a table of comparing results.Simulation experiment shows that the recognition rate of proj ectile target significantly improved compared to recognition relying on one single feature.