Research On Improved Approaches Of Precision And Efficiency Of Particle Filter Track-before-detect Algorithm

At present,targets are often submerged in a lot of noises and will have low signal-noise-ratio as a result of the diversification of targets and the complexity of environment in the actual tracking system.In this case,even though a target appears,due to its signal is too weak,it may not be detected in a scanning.Because of that,traditional detection and tracking technology encounters great challenges.The Track-Before-Detect(TBD)technology could process raw data directly and integrates the detection and tracking processes.This characteristic makes the TBD technology perfectly suitable for detecting and tracking weak targets with low signal-to-noise ratio.The TBD technology is a typical strong nonlinear problem and the Particle Filter(PF)algorithm is an effective solution for nonlinear filtering.Therefore,the Track-Before-Detect based on Particle Filter(PF-TBD)algorithm has been designed and it becomes an important way of exploring weak targets effectively.More researches about it should be done to make progress in this filed.The main thesis of this research is PF-TBD algorithm and the main research are as follows:1.A particle filter based on hybrid optimization sampling(HPF)algorithm is first proposed in this thesis,which against the low detection and tracking performance of the PF-TBD algorithm and the particle collapse problem,that are brought from the particle impoverishment phenomenon,which is caused by the particle update process and the resampling process.The HPF algorithm introduces the hybrid optimization algorithm,which combines the Simulated Annealing(SA)and Differential Evolution(DE),into PF algorithm.In this way,the resampling is regarded as the process of seeking the optimal solution of objective function of hybrid optimization algorithm.And the diversity of particles is increased.Therefore,a high-precision paricle filter track-before-detect algorithm based on hybrid ooptimization is further proposed.In generic PF-TBD algorithm,particles are updated by two parts separately,one part is birth particles,and the other part is continuing particles.The prior density function is chosen as the importance density function of birth particles.The states of continuing particles are usually obtained from the state equation.In this case,the update process does not take into account the amendment effect of the latest particle information.And then,the particle impoverishment degree after the resampling process will become aggravated.The sampling particles will be far away from the true posterior probability distribution of the target state,which will affect the detection and tracking performance.Therefore,the proposed algorithm introduces the hybird optimization idea into PF-TBD algorithm.In this algorithm,the updated particle set is optimized by hybrid optimization operator and the optimal updated particle set is generated.The proposed algorithm reduces the impoverishment problem degree of the resampling process and makes the estimated particle state values are closed to the true particle state values.Simulation results show that the proposed algorithm is available and its performance of detection and tracking weak targets is superior to three PF-TBD algorithms with three basic resampling methods.2.A Lamarckian particle filter(LPF)is researched against the low utilization of paritlces and the heavy computation problem,which is caused by intelligent optimization sampling methods.The performance of PF-TBD algorithm is directly affected by the performance of PF algorithm.The optimization sampling methods have good results to solve the particle impoverishment problem and increase the estimated precision.However,these intelligent optimization PFs also have some deficiencies,such as controlling the particle diversity,guiding the optimization process and so on.Moreover,these algorithms have greatly increased the algorithm complexity,correspondingly increased the computation and running time and limited its applicability in practice.Based on this,the proposed algorithm introduces Lamarckian inheritance optimization idea into PF,which is different from genetic algorithm.A new optimization sampling strategy,which combines the overwriting operator and the elitist preservation strategy,is structured to achieve resampling process and the optimal particle set will be chosen.The overwriting operator is designed by Lamarckian inheritance theory.The propsed algorithm regards every paritlce as the population of the optimization sampling process,every bit of every particle as a individual and obtains the optimal particle set.The proposed algorithm not only increases the diversity of particles,but also simplifies procedures of generic intelligent optimization methods.In addition,a Lamarckian particle filter based on unscented Kalman filter(ULPF)algorithm is proposed,which improves the importance distribution choice of LPF algorithm.Simulation experiment results show that both the proposed algorithm and its improved algorithm not only reduce the computation and the runtime,but also improve the filtering estimated precision,increase utilization of particles.3.An efficient particle filter track-before-detect algorithm with Lamarckian inheritance strategy is presented against the required more particles and low execution efficiency problem.The PFs based on intelligent optimization sampling ideas indeed improve the performance of detecting and tracking weak targets of PF-TBD algorithm.However,the algorithm implementation will lead to increase the computation.Meanwhile,the number of particles of PF-TBD algorithm is proportional to regression level of probability density function.So,many particles are needed to ensure the tracking accuracy of PF-TBD algorithm.These will lead to intelligent optimization sampling PFs with high precision performance cannot be used in engineering applications in the future.The proposed algorithm utilizes the Lamarckian inheritance optimization sampling method to design the particle update process of PF-TBD algorithm based on the efficiency of simplifying PF algorithm of Lamarckian inheritance idea.This update strategy adopts each floating-point value of particles to optimize algorithm.The proposed algorithm improves utilization of particles,which means that it can use fewer particles to reach corresponding detection and tracking precision.And the proposed algorithm also reduces the computational and runtime.Simulation experiment results show that the proposed algorithm is feasible and effective,and not only improves detection and tracking performance,but also in its utilization of particles and runtime.