SAR Target Detection And Discrimination Based On Polarimetric Information

Polarimetric Synthetic Aperture Radar(Pol SAR) alternately transmits and receives radar signals in different polarizations, and can obtain abundant information about target scattering, which plays an important role in feature extraction, Image Interpretation Technology and Autio Target Recognition. This thesis aims at the extraction and using of polarmetric information, and focuses on several key problems, including Pol SAR target decomposition,Pol SAR target detection and discrimination.In the beginning, we will briefly introduce the research background and significance of Pol SAR target decomposition, target detection and discrimination. And based on this, the main content of this dissertation will be introduced in three parts in detail.Firstly, the paper introduces the basic theory on electromagnetic polarization information processing. It introduced one of the most important preprocessing of polarimetric data i.e, polarimetric calibration. This part studies the calibration based on single target and based on distributed target. And we analyze the influence of calibration for relative phase, system cross talk and the system channel imbalance.Secondly, the paper studies the topic of polarimetric target decomposition, and mainly focusing on the model-based decomposition. This work is studied in following two aspects. 1), Several traditional algorithms of model-based decomposition is discussed, including Freeman-Durden decomposition,the Improved Yamaguchi Decomposition, Model-Based Decomposition Constrained for Nonnegative Eigenvalue, Four-Component Scattering Power Decomposition With Unitary Transformation of Coherency Matrix and so on. 2), A scattering power decomposition method based on polarimetric similarity of coherency matrix is proposed. By estimating the largest polarimetric similarity between the original cohenrency matrix and the basic scatteri ng mechanism, the dominant scattering mechanism can be get,then the dominant scattering energy can be extracted under the constrained of semidefinite, which avoid the under estimation of the dominant scattering energy and get a better decomposition result.Thirdly, the paper studies the target detection and discrimination based on polarimetricinformation. At the stage of target detection, the traditional two parameter constant false alarm rate(CFAR) detector is mainly studied. To deal with the problem for CFAR detector under complicated scene, this paper studies a polarimetric detection algorithm based on polarimetric information and support vector data discription(SVDD). SVDD constructs a tight hyper sphere boundary around target data and shows good property for the one-class classification problem. By extracting large amount of polarimetric features and training a good suppert vector plane using SVDD,the test samples are classified and a binary image representing target and clutter can be obtained. By using morphological filter to the binary image, finally, the target chips can be got. At the stage of target discrimination, the thesis introducs some traditional features for discrimination developed by Lincon Laboratory and some other Laboratories. And their porformences are in detail by experiments. For the discriminator, Gauss discriminator and SVDD discriminator are introduced and used for the target discrimination.