Image Processing And Application Of Polarimetric Synthetic Aperture Radar

At present, polarimetric synthetic aperture radar is an important aspect of the microwave imaging research and application, which can efficiently enhance the ability to acquire the information of targets, and provide an effective approach to further detect, identify and class targets. Some problems of polarimetric SAR were studied and explored in an all-round and systematical way, which include the basic theories, polarimetric information processing and application. In condition, some new concepts and algorithms were presented.Firstly, the polarimetric description of electromagnetic Waves was introduced, some methods of the polarimetric description were provided, and the basic theories of polarimetric measurement were studied, the concepts of scattering coordinate system and polarimetric scattering matrix were introduced. All above laid a foundation for the research of polarimetric information processing and application. Then, target optimum polarization and polarimetric synthesis technique were studied. By constraining polarimetric states of transmitting and receiving antennas, co-polarized and cross-polarized signatures were obtained. Because the received wave was partially polarized, completely polarized, completely unpolarized, and total available power signatures were defined based on the power of the return wave. Target optimum polarization could be easily obtained by searching on the signature space, and these optimum polarizations could be used in target identification and classification. Some general analysis methods of polarimetric SAR image were also studied. In order to improve the contrast degree of two targets, the polarimetric contrast optimization was researched.In polarization image filtering, polarimetric whitening filter, optimum weighing filter, multi-texture maximum likely estimation, and local statistical filter were studied. Their performances of different filters were evaluated by using the waveform of sample signals, polarization signatures and the relative standard deviation. In polarization target detection, by combining gray value with shape of target, artifical targets were detected based on extended fractal. In polarization target identification and classification, support vector machine classifier was studied, classification experiments were implemented by applying it to measured polarimetric SAR data.The innovations of this dissertation are as follows:(1) A new algorithm of obtaining target optimum polarization was presented. Five signatures were defined based on polarization synthesis technique, then target optimum polarization could be obtained to partially polarized wave by searching on the signature space. This algorithm was convenient because of avoiding complex mathematic deduction. Considering the speed of calculation, the polarimetric states of transmitting and receiving antennas should be constrained. Thus these optimum polarizations were a local solution, not an global solution.(2) Because the polarimetric stares of transmitting and receiving antennas could be depicted by the ellipticity angle and the orientation angle, and these angles could be used in target identification and classification. A new algorithm of target classification was presented based on the ellipticity angle and the orientation angle. Efficiency of this classification algorithm is good because we can select target optimum polarization of bigger change to different targets.(3) Extended fractal integrates gray with shape. It can detect targets by measuring the roughness degree of texture. Two schemes of target detection were presented by combining extended fracta] with the polarimetric invariant. These schemes could detect artifical targets, for example ships, bridges, aircrafts, tanks.(4) A new algorithm of evaluating the efficiency of the polarimetric filter was presented based on the comparison of different filters and the validation of experiment data. The resolution of image was evaluated by the waveform of sample signals, the polarimetric information was evaluated by polarimetric signatures, the speckle reduction was evaluated by the relative standard deviation. The results indicated local statistical filter had an effective performance because it could reduce speckle and preserve the resolution of image without damaging polarimetrie information.(5) Many polarimetric target decompositions were explored, and support vector machine was applied to the classification of polarimetric SAR image. A new algorithm of target classification was presented based on polarimetric target decomposition and support vector machine. Finally, the effect of classification results was discussed by selecting different kernels and different parameters. The results indicated this algorithm was very efficient to classify targets and it could improve performance of classification by selecting appropriate parameters.In conclusion, some problems of polarimetric SAR information processing and application were studied, some new concepts and algorithms were presented, many research results were obtained. I hope this dissertation is helpful to system designers and data analysts of polarimetric SAR.