Research On Particle Filter Algorithm For Radar Target Tracking Before Detection

With the diversification of targets and the complexity of the marine environment,the current marine radar is facing a very serious challenge,and the detection and tracking of weak targets is one of them.Track-Before-Detect(TBD)is an effective method to detect weak targets in low SNR environment.Particle Filter Track-Before-Detect(PF-TBD)is one of the main ways to realize TBD technology.PF-TBD algorithm has the characteristics of recursive processing in nonlinear and non-Gaussian application scenarios.The implementation process is simple and the accuracy can approach the optimal estimation,which is a research hotspot in recent years.This thesis mainly studies the radar weak target PF-TBD algorithm,the main contents are summarized as follows :(1)The basic principles of two radar signal processing methods are summarized,and their applicable scenarios,advantages and disadvantages are compared.The Bayesian recursive estimation framework is summarized,and the particle filter algorithm,one of the approximate algorithms,is deduced in detail.Aiming at the problem of particle degradation,three solutions are given : increasing the number of particles,introducing resampling and selecting the appropriate importance density function.Finally,Bayesian estimation under multi-objective conditions is analyzed,which lays a theoretical foundation for subsequent work.(2)Joint detection and tracking for a single weak target.Two kinds of particle filter TBD algorithms are compared and studied : Standard Particle Filter Track-Before-Detect(SPFTBD)with mixed estimation of target state and existing variables and Efficient Particle Filter Track-Before-Detect(EPF-TBD)with separate estimation of target state and existing variables.The EPF-TBD algorithm is proved to be superior to SPF-TBD algorithm in terms of detection performance,estimation accuracy and computational efficiency.Then based on the EPF-TBD algorithm,the influence of different particle number,different target influence resolution unit number and different resampling methods on the performance of the algorithm is analyzed from the aspects of accuracy and operation efficiency,which provides a reference for setting the filter parameters reasonably.Finally,aiming at the problem of low particle utilization in EPF-TBD algorithm,a method to improve the importance probability density function of new particles is proposed.Based on prior information,low-threshold uniform sampling is used to generate new particles.Simulation results show that compared with uniform sampling,EPF-TBD algorithm based on low-threshold uniform sampling has higher detection probability and tracking accuracy.(3)For the joint detection and tracking of multiple weak targets,the standard particle filter pre-detection tracking algorithm with known target number and the particle filter probability hypothesis density pre-detection tracking algorithm with unknown target number are studied.Firstly,based on the single target TBD model,a TBD model for multi-target scene is established.Then,according to the number of targets and the distribution of targets in the observation scene,the multi-objective SPF-TBD algorithm is used to study the problem of multiple weak targets with known and time-varying number of targets.PF-PHD-TBD algorithm is studied with probability hypothesis density(PHD)when the prior information of target number is unknown.Through the optimization of the measurement model and the Poissonization of the noise,the PF-PHD filter that can be applied to the TBD scene is given,and the advantage that it does not need complex data association in dealing with multiobjective problems is fully exploited.The simulation results show that the algorithm can accurately estimate the number of targets in the case of unknown number of targets,and has good tracking performance.