모노스태틱/바이스태틱 HRRP를 이용한 편대비행 표적식별 연구

Abstract

High resolution range profile (HRRP) is 1-dimensional RCS distributions that can be generated through radar reflection signals and provide very effective radar signature for target recognition using radar. Since HRRPs vary considerably depending on the aspect angle, single target classification is performed by training HRRP according to observation angles in the database. However, when there are many targets in the single radar beam, HRRPs are generated in real time by the type and the number of targets, real time projection position, and scale factor according to variance of radar beam. In addition, since it is impossible to predict the above parameters in advance, unlike the single target identification problem, it is difficult to perform the multiple targets identification using the previously trained database. This paper proposes an algorithm that can estimate the number of formation flight using monostatic and bistatic high resolution range profiles (HRRPs). Using the RELAX Algorithm, single and multiple targets are distinguished and HRRPs are reconstructed. In addition, the MUSIC Algorithm and CFAR detector are applied to obtain distinct scatterers of each target of the multiple targets. In order to overcome the limitations of the monostatic radar vulnerable to targets flying in the line-of-sight direction, the data from the bistatic radar were fused so the number of targets were successfully estimated regardless of the signal-to-noise ratio variation. And HRRPs of a single target that have been trained previously are combined and then a real time database is constructed to perform multiple targets classification. The projection position and the scale factor are optimized and combined through particle swarm optimization (PSO), and the process is repeated according to the type and number of targets to construct a real time database. Experimental results were obtained at SNR = 10, 15, 20, 25, 30dB. HRRPs of targets flying in formation were generated by randomly selecting each parameter within a specific range. In order to reduce measurement error caused by additive white gaussian noise (AWGN), simulation was repeated 50 times in each SNR environment. In the environment with SNR of 10dB or more, 100% classification result was obtained.