Study Of Variant Target Recognition Based On Structured Sparsity Prior

High-resolution range profile(HRRP)contains abundant structure signatures of the target,and is easily obtained and processed.Thus HRRP based recognition method is received intensive attention.When the weapon type and configuration of target varies,the HRRP will also change and generate a lot of variant forms which deteriorates the recognition performance of HRRP based method.Thus the robust identification of the variant form of the target needs to be realized.As the variant structure exists continuously in the space and only affects the local part of the target,its effects on original HRRP(variant component)show a structured sparse feature.In this paper,we use this kind of feature of the variant component to model the variant component separately,separate the variant component,and recover the HRRP of the original target.Use the recovered HRRP for recognition to improve the performance of variant target recognition.The main content is summarized as follows:1.Using electromagnetic simulation software to simulate radar echoes of two kinds of CAD model.According to the analysis of these radar echoes,we find that in partial attitude angles the variant component is local and continuous existence on the original HRRP.Considering this feature is similar to the structured sparse feature of sparse signal processing,so we review the sparse representation theory and structured sparse model briefly.2.The variant target recognition is realized under the constraint optimization framework.Firstly,we introduce a structured sparse model based on coding complexity.This model can reconstruct the structured sparse signal without any prior information about the block length and number.Secondly,we model the variant components with the above model and use the training sample to represent the original HRRP,propose a structured sparsity based HRRP occlusion model.After the model inference,we separate the variant component.Experiment results show that the recognition performance can be improved effectively with the recovered HRRP.3.The variant target recognition is realized under the Bayesian framework.Firstly,we introduce a hierarchical Bayesian model based on pattern coupled prior and describe the model inference process in detail.This model can promote block sparse property of the signal.Secondly,we model the variant components with the above model and use Sparse Bayesian Learning(SBL)to describe the original HRRP,propose a local pattern coupled based HRRP occlusion model.After the model inference,we separate the variant component.Thirdly,we analyze the computational complexity of the model briefly,and then reduce the computational complexity by compressing the inverse matrix dimension.Experiment results show that the recognition efficiency of the algorithm is improved,and the performance of the algorithm is not affected.