Research On Extended Target Tracking Algorithm Based On Random Finite Sets In Passive Sonar

At present,passive sonar detection systems based on unmanned small-size platforms are widely used in underwater environment and noise measurement,aquatic species ecological research,regional early warning and anti-submarine,etc.But due to the limitations of bandwidth and speed,it is difficult to transmit complete raw data back to the shore base.It is a feasible option to use certain data processing methods to obtain information from raw data and then transmit it.At the same time,due to the data processing needs of a single platform,multiple targets tracking has become an extremely significant task.Considering that the number of targets in the surveillance area changes frequently and the scene is complicated,the multiple targets tracking method based on Random Finite Sets(RFS)has obvious advantages,and the target radiation noise is characterized by multiple measurements under passive sonar observation.Therefore,under the above background,this paper carried out the research on extended target tracking algorithm based on RFS in passive sonar.Firstly,studied the RFS based multiple targets Bayesian filter and the realization of its firstorder moment approximation form,Probability Hypothesis Density(PHD)filter,under the passive sonar bearings-only data.Researched the multiple targets system model based on RFS and the optimal multiple targets filter under Bayesian theory,and carried out research on the Gaussian Mixture PHD(GM-PHD)algorithm and related performance evaluation methods.The ability of GM-PHD algorithm to deal with complicated scenes with new and dead targets is verified by processing simulation data.Secondly,aiming at the fact that the target radiated noise signal has a bandwidth,which makes it have multiple measurements under passive sonar observation,the PHD algorithm research based on extended target modeling is carried out.Researched and analyzed that the azimuth observation of the target under the Direction-of Arrival(DOA)method of passive sonar cross-spectrum method belongs to the extended target type,and built an extended target model based on the characteristics of the target and data in passive sonar observation,combined with the GM-PHD method to realize the Extended Target GM-PHD(ET-GM-PHD),and based on the simulation data to study the azimuth intersection problem under the extended target,an improved subpartition method that introduces frequency as auxiliary information is proposed,and the method is tested in the sea test data.Finally,aiming at the unstable performance of the PHD algorithm in scenes with high clutter density,the Extended Target Cardinality PHD(ET-CPHD)algorithm is studied.Researched the processing idea of CPHD algorithm and its implementation in DOA,and studied the "spooky action" in CPHD when the target missed detection and the dynamic reweighting method to solve this problem.Through the simulation data,the processing performance of ET-GM-CPHD under high clutter density and the processing effect of the dynamic reweighting method under the passive sonar data are verified,and the sea test data is processed to further verify the algorithm in the actual environment processing capacity.