Research On Radar HRRP Target Rejection Methods

Radar automatic target recognition(RATR)can provide information such as the type and model of the target.It has important application value in the military field and has attracted widespread attention from researchers in the radar industry.High resolution range profile(HRRP)can be obtained easily.And it needs small storage.So it is an important feature in the application of RATR.In the practical application background,it is necessary to identify non-cooperative targets and even hostile targets.It is difficult to build a database of these outlier targets.So the training sample database cannot be complete.For new test samples,if it is an outlier target,it is unreasonable to judge it as any type in the database.Therefore,for the RATR system,the outlier targets should be rejected first during the identification process,and then classify the samples in the database.This dissertation mainly focuses on the object rejection technology outside the library,based on traditional models and deep networks.The main research contents of this article include three aspects.1.The basic theory of radar HRRP target rejection is introduced.Firstly,this paper introduces the commonly used HRRP features,including time domain features,spectrum features and spectral characteristics.Then it introduces the commonly used rejection algorithms.Finally this paper introduces the commonly used indicators of rejection performance evaluation.They provide quantitative indicators for the evaluation model reject performance.2.To improve the rejection performance of the traditional support vector data description(SVDD)model,this paper proposes a bayesian support vector data description(BSVDD)model.The BSVDD model uses augmentation technology by introducing hidden variables on the SVDD model.And it assigns a reasonable prior distribution assumption to the SVDD model parameters.The SVDD objective function is replaced by a probability model to realize the combination of traditional SVDD model and bayesian frame.The results of simulation experiments show that the method of rejecting outlier targets based on bayesian support vector data description effectively improves the rejection performance of the system.3.HRRP data has the characteristics of high dimensions and complex spatial distribution.Using traditional SVDD model for rejection has the problems of selecting the kernel function form and optimizing the nuclear parameters.Moreover,using the SVDD model requires a lot of memory to store support vectors,which limits its application in high-dimensional data scenarios.In view of the above problems,a deep support vector data description(Deep SVDD)model is applied to the rejection task,and the rejection performance of the model on different features is analyzed in detail.The Deep SVDD model uses deep networks instead of kernel functions to implement non-linear mapping of features.Compared with the conventional kernel function,the learned network can map it to a more separable feature space according to the characteristics of the HRRP data,which helps to improve the rejection performance.The Deep SVDD model is designed using convolutional neural network(CNN)and long short-term memory(LSTM),and the rejection performance is tested on the one-dimensional HRRP spectral amplitude characteristics and two-dimensional HRRP spectral characteristics.Simulation results show that the rejection method based on deep support vector data description for outlier target rejection is superior to traditional rejection methods.Moreover,The rejection performance of the Deep SVDD model using LSTM is better than CNN.