Research On Intelligent Processing Methods Of Polarimetric Synthetic Apeture Radar Image

With the development of polarimetric SAR, the spatial and temporal resolution of the data continue to increase, thus resulting in a sharp increase in the amount of data. For this reason, how to make effective use of polarimetric SAR data to extract and interpret information of the scattering mechanisims is a important issue. As the main objective of the polarimetric SAR image processing, intelligent processing has been playing an important role in the field.In this paper, some comprehensive and systematic researches on polarimetric SAR image processing using intelligent optimization algorithms and kernel method have been done. Firstly, the fundamental theory of electromagnetic wave in polarimetry radar, including the polarization representation of electromagnetic waves, polarimetric scattering description, and target polarimetric scattering characterization has been introduced. Then, a brief introduction of the methods such as the classical evolutionary algorithms-Genetic Algorithms, the representation of swarm intelligence algorithms-Particle Swarm Optimization and the kernel methods has been given.The content of the research in this paper includes three aspects:polarimetric SAR speckle reduction, polarimetric contrast enhancement and polarimetric SAR data classification. The post-processing methods for speckle reduction vary from spatial domain to polarimetric domain. The mainly discussed methods were spatial domain speckle reduction method including mean, median and statistical filtering, polarimetric whitening filtering and optimal weighting filtering. However, the method based on kernel principal component analysis was the key method. For polarimetric contrast enhancement, A PSO-based method has been proposed after discussion on the classical methods, the optimization of polarimetric contrast enhancement method and the Co-Null method. The current polarimetric SAR data classification methods can be divided into several kinds, most of them are based on statistical characteristics, physical scattering properties, and others are based on both the statistical and physical characteristics. In this paper, some combination of H/αa/A and H/αpolarimetric SAR intelligent classification algorithms have been proposed, the efficiency and accuracy of the method also have been analyzed.The innovations of this dissertation are as follows: (1) The classical spatial filtering method usually couldn’t take into account both speckle reduction and edge preservation while the polarimetric method can not make full use of the polarimetric information.Thus, a filtering mothed based on kernel independent component analysis has been proposed. To obtain an enhanced SAR image with low speckle, the target information in polarimetric data were used as much as possible to reduce the speckle and all channels were involved in processing. The proposed algorithm can effectively use information in the four polarization. Experiments showed that kernel independent component analysis algorithm has better ability of achieving good filtering effect and maintaining the edge information compared with other filters.(2) A polarimetric contrast enhancement method based on particle swarm optimization algorithm has been proposed with the "three-step" model which simplify the parameter solution. Moreover, the method avoids the mathematical complexity of traditional optimization methods, and has the characteristics of high contrast. Compared with the genetic algorithm, the method incorporates the features of simple computation and convenient operation.(3) Intelligent polarimetric SAR classification based on H/a/A and H/a was proposed. The main work in this part is the improvements of particle swarm optimization algorithm, these improvements include:increase the convergence speed by making dynamic change in the number of particles, balance both global and local search capabilities and achieve faster convergence speed by changing inertia weight. Intelligent algorithm for polarimetric SAR classification not only overcomes shortcomings of being trapped into local optimum and sensitive to noise in the traditional H/α/A-Wishart classification, but make full use of the physical scattering and statistical characteristics of polarimetric SAR data, and it also has the advantages of simple implementation, powerful search capabilities and achieving higher classification accuracy.