Study On HRRP Generation Based Unknown Target Discrimination

In the field of Radar Target Recognition, High Range Resolution Profiles (HRRP)is increasingly becoming an important means of recognition because it’s obtained easilyand can be recognized in real time. In the traditional way of recognition based on HRRP,the training samples of the to-be-recognized target are needed in order to get a correctresult. When encountered an unknown target, the traditional way of recognition can’t beused because of the lacking of training samples. Such case usually leads to error results.This paper conducted an in-depth research in this problem, the main content is asfollows:1. To solve the problem of unknown targets’ lacking training samples, the paperprovides a method to generate the HRRP base on the Gamma model. Little priorknowledge of the unknown target is used to estimate the parameters of the Gammamodel in this method and then the training samples of the unknown target are generated.The results of the experiments show that compared with Gaussian model and Uniformmodel, the training samples generated by Gamma model are closer to follow thestatistical distribution of the unknown target’s HRRP.2. Discrimination results are affected by the single kernel function which is used inthe traditional Support Vector Machine (SVM). In order to solve the problem, a methodof mixed kernel function based Support Vector Machine is studied. Since thePolynomial kernel function has a good generalize ability while the RBF kernel functionhas a good learning ability, a new mixed kernel function is created by combining theabove two kernel functions so that it has the advantage of the both. The simulationresults show that the mixed kernel function based SVM can improve the discriminationresults compared with traditional SVM.3. Data sets of known target are required to follow spherical distribution in thetraditional Support Vector Data Description (SVDD). However, it’s difficult to meetsuch condition in practice. In order to solve the problem, a method of cluster basedSVDD is studied. The data sets of the known target are divided into several groups by cluster at first and then SVDD is used in each group to get a discrimination result. Sincethis method takes into account the distribution of the data sets and several SVDD areused rather than one, better discrimination results can be gotten. Meanwhile a parameteroptimization method is proposed to solve the problem that parameters are difficult to beset in the support vector based single class discrimination. The simulation results provethe feasibility and effectiveness of the above methods.