Broadband Polarization Radar Target Recognition

With the development of wide-band technology and the improvement of polarimetric technology, radar becomes able to get high resolution range profiles and fulfill full polarization measurement or polarization shifting, is becoming the mainstream of modern radars, which provide new thought and method for radar target recognition. Combination of high resolution range profiles with full polarization may become the most hopeful approach to resolve the problem of radar target recognition, and thus it becomes one of focused methods for automatic target recognition.This dissertation systematically studies the theories of wide-band polarimetric radar targets recognition. First we get scattering data of wide-band polarimetric radar plane targets, which have similar shapes and sizes, from computing polarization scattering matrix of typical ccated targets, and then study every stage of wide-band polarimetric radar target recognition: In the preprocessing stage, by using a new theory of machine learning, a novel correction algorithm based on Kernel-based Nonlinear Representor (KNR) is presented for radar reciprocity, which provides a new approach for correction of radar reciprocity; In feature extraction, Fractional Fourier Translation, Target Decomposition and polarized invariants generation are adopted to extract features from wide-band polarimetric radar targets. Fractional Fourier Translation is widely used in time-frequency analysis, while Target Decomposition and polarized invariants are used in narrow-band polarization radar target recognition. In classification stage, five kernel-based classifications are used and compared, and fusion methods are designed for wide-band polarimetric radar target classification.Reciprocity correction experiments on random scattering matrixes show that correction algorithm based on KNR is more effective than other algorithms. Experiments on scattering data of five planes show that all the feature extracting algorithms are effective, and extracting algorithm based on Fractional Fourier Translation has the highest recognition rate. In all of the kernel-based nonlinear classifications, KNR has the best comprehensive performance. In all of the fusion algorithms, D-S algorithm has the highest recognition rate.