Research On Recognition Of Real And Pseudo Warheads Based On Deep Learning

In a background of modern war and multi-sensor cooperation,the ability to effectively recognize the real warhead of ballistic missiles via information technology is an important prerequisite of improving the effectiveness of anti-missile system and consolidating national defense.Artificial intelligence is introduced into the recognition of radar targets in ballistic midcourse,especially the real and pseudo warheads,which can complement the classical signal processing method and overcome the bottleneck of recognition rate of the latter under different observation conditions.This paper is theoretically based on the difference of dynamic characteristics between real and pseudo warheads,and takes deep learning as the main technical approach to carry out a series of research:1.The motion pattern and parameter calculation method of real and pseudo warheads are studied.The micro-motion feature of cone targets under the influence of external torques is discussed.Based on the principles of rigid body mechanics,the relationships of precession angle,precession period and target weight are derived.2.For cone warhead,rigid copy decoy(pseudo warhead),ellipsoid balloon decoy and three kinds of heavy decoys with different designs,the target model is established,their dynamic HRRP sequences are generated,utilizing the FEKO electromagnetic simulation software.3.The method of using deep learning to solve the ballistic target identification problem,especially the recognition problem of real and pseudo warheads,is studied.Two network structures for processing HRRP sequences and other time-domain related twodimensional signals to conduct target identification is proposed and tested: one completely based on CNN,the other based on CNN and LSTM.4.The positive performance of multi-sensor data fusion in improving the stability and effectiveness of anti-missile system is discussed.The advantages of multi-sensor data fusion basing on the network structures proposed in this paper is verified through simulations.