Radar Target Recognition Based On Kernel Methods And Manifold Learning

Target recognition is a terminology contains plentiful information ,it implicates extracting worthful information and stable feature of target from electromagnetic scattering .In the condition of radar irradiate, feature of target electromagnetic scattering wave contains: motion parameter, target shape, measurement, engine and propeller modulation ,polarization and so on. These feature which are the base of radar target recognition indicate target property from different aspectRATR using high resolution range profiles (HRRPs) is researched in this disertation. An overview of the background and significance of RATR is shown first. Then, feature extraction and classification techniques in HRRP-based RATR are summarized and studied.Based on the scattering center model, we analysis the electromagnetic scattering characteristics and scattering mechanisms of aircrafts, and the formation mechanism and characteristics of HRRPs. HRRPs of five aircraft categories within a certain attitude angle range are simulated. Next, some preprocessing methods including distance alignment and normalization are summarized.We use two types of methods in feature extraction. One type is the subspace methods, including principal component analysis (PCA), linear discriminant analysis (LDA), kernel-based principal component analysis (KPCA) and kernel-based fisher discriminant analysis (KFDA), and the other one is manifold learning, that is Local Linear Embedding(LLE),Local Preserving Projection(LPP),Neighborhood Preserving Projection(NPP). Whereafter, several classifiers including the Euclidea distance, Radial Basis Function neural network (RBFNN), Support Vector Machine (SVM) are adopted and compared.Experimental results on both simulated and measured HRRPs show comparatively good performance of the techniques.