Study Of Variant Target HRRP Recognition Based On Bayesian Structured Statistical Model

In the modern world,information is becoming more and more important.RADAR's tasks are not only simply target detection and ranging,but also getting more information about the target for us to master the enemy's opportunities.Therefore,radar automatic target recognition is drawing more and more attention.Because RADAR high resolution range profile contains the structural information needed for target recognition,and is easy to acquire and calculate,HRRP is often used for radar automatic target recognition.HRRP changes according to the structure of the target,which varies because of target's different loads and tasks,resulting the HRRP mismatch with the model,thus lower the recognition performance.In this thesis,we studied the problem that the recognition performance is degraded due to the model mismatch caused by the variant target's structure difference.The recognition performance is improved by removing the variant signal component that causes the model mismatch.The main work of the paper is summarized as follows :(1)Through the analysis of the electromagnetic scattering center model and the observation of the simulating RADAR target echo,it is found that the components causing the model mismatch in the variant target signal have the characteristic of structure sparsity.Therefore,a sparse representation model is established for the variant target signal.The algorithm,which is widely used in the compress sensing area,is used in this work to separate the variant component from the variant target signal,and then using the HRRP without variant component to recognition;(2)Respectively,using the Markov chain hyper-prior Bayesian structuring model and the non-zero parameter coupled Bayesian structuring model to assure the structure sparsity,establish a local Markov chain hyper-prior Bayesian Structural model and local non-zero parameter coupled Bayesian structuring model.Both models can solve the local structure sparse representation of the variant target signal,obtain the variant component and remove it from the variant target signal,then use the variant component removed signal for target recognition.The recognition performance will be improved because the component causing the model mismatch has been removed;(3)The Bayesian statistical model solving process consume a lot time in computation.Thus a generalized approximate message passing algorithm is introduced to replace the traditional probability derivation method to solve the mean and variance of the sparse representation.The GAMP-accelerated Bayesian sparse recovery algorithm can separate the variant component from the variant target signal and then recognition,which not only improves the operation speed,but also does not cost much recognition performance sacrifice.