Research On The Ship Target Recognition Based On High Resolution Range Profiles

With the development of wideband radar, the resolution of radar is much smaller than the size of the target. Target echo can be viewed as the superposition of the echo of multiple scattering points at diffe rent reflected intensity. Wideband radar can not only detect the location of the object, but also obtain the structural information of the target. It urges the rapid development of the technology of Radar Automatic Target Recognition(RATR) and turns into one of the most important research areas in the field of military. High Resolution Range Profiles(HRRP) is obtained from the wideband radar based on the scattering center model consisting the target location and structural information such as the size and material quality of the target. Due to the simplicity of obtaining the HRRP and its real-time performance, the RATR based on HRRP goes in-depth study.This paper studies recognition technology which is based on ship target’ HRRP, analyzes the property of HRRP theoretically through the establishment of ship model and processes stability of HRRP through the envelope a lignment and incoherent average. Radial length of target has a clear physical meaning and can reflect the basic attributes of the target. This paper pretreats data of actual measurement, studies the different methods to extract target radial length in order to find the optimal extraction method.However, it is not reliable to recognize and classify the sensitive attitude change of HRRP by exploiting the length of the feature. It is crucial to extract the robust and representative characteristic information of HRRP. Power spectrum of HRRP has the property of invariance when translating, but cannot deal with non-gaussian noise effectively. While higher order spectrum especially double spectrum which not only has an attribute of translational invariance and also can handle the non-gaussian noise better was introduced into the field RATR. Therefore, double spectrum has been widely applied in the field of RATR. But the dimension of the double spectrum is extremely high and requires higher storage and computation of radar system. Therefore, dimension reduction of the double spectrum has become important research direction. This paper analyzes and simulates four design features of dimension reduction double spectrum using Support Vector Machine(SVM), and different parameter optimization algorithm is adopted to study the identifying performance and influencing factors of the dimension reduction of double spectrum. At last, it contrasts the difference of the identifying performance between the SVM and Artificial Neural Network(ANN)Because HRRP based on ship target is a kind of nonlinear structure data, classification and identification by means of HRRP endures not only a dramatically large amount of computation but also a relatively low rate of identification. Feature subspace method makes projection from the original data to feature subspace through the nonlinear conversion and extracts the eigenvectors which represents most of the original data in feature subspace to make it have linear separability. This paper studies the Combination Kernel Principal Component Analysis through analyzing of Principal Component Analysis(PCA) and the principle of Kernel Principal Component Analysis(KPCA), and proposes a joint algorithm of PCA, KPCA and LDA taking the advantages of linear discriminant method(LDA). Additionally, simulation data is used to study the influence of the training sample and the attitude angle change on extraction methods of different features and verifies the methods by testing the measured data.