Study On Several Problems In Polarimctric SAR Processing

Early synthetic aperture radar (SAR) often used a single polarization of transmittedradiation, for example, transmitting horizontally polarized radiation and receiving horizontalpolarization (single-pol systems). Fully polarimetric systems (quad-pol systems), whichalternately transmit two orthogonal polarizations and record both received polarizations, allowfor much more information about the target scattering characteristic to be extracted from theradar signal comparing to the single-pol systems. The polarimetric information can berecorded in the form of the complex scattering matrices, with which the scatteringcharacteristic of the targets can be extracted and analyzed. Because of providing the wholescattering characteristic of the targets, the polarimetric SAR (PolSAR) is of great importancein the remote sensing, which can be used widely in both civil and military fields. And thisdissertation focuses on the polarimetric SAR application, which makes a study on exploitingthe useful information in PolSAR for the remote sensing application. The primarycontributions of the dissertation are as follows:Contributes to the RGB model, a pseudo color fusion image of PolSAR based on thesame normalized factor can be achieved. However, since the power of echoes is13dB largerin a co-polar channel than in a cross-polar channel, the same normalized factor will lead to theunbalanced RGB distribution. To solve this problem, a method of choosing differentnormalized factors for different polarimetric channels is proposed. Meanwhile, since differentpolarimetric channels have different dynamical ranges, a feature extraction methodnonnegative matrix factorization (NMF) is introduced and used to extract the primary featureof the data. By matching the histograms of the span to the extracted feature, the reconstructeddata reconstructed by using the matched feature is employed to carry out the pseudo colorfusion. Experiment with measured data validates the effectiveness of the proposed methods.When implementing the speckle reduction in PolSAR image, the detail information ofthe boundary and the scattering characteristics of targets can hardly be preserved. To solvethis problem, an improvement to Lee filter based on the product speckle model is studied.Making using of H/α decomposition, the different sorts conforming to the inherent physicalscattering characteristics can be marked off with different markers (pixels belonging to thesame sort are marked with the same marker). Referring to the different markers, an adaptiveslipping window is designed to estimate the filtering parameter for the central pixel of thewindow. Measured data are applied in the filtering experiment using the estimated filteringparameters, and the capacity of the proposed method is confirmed. Fully polarimetric real data from China is studied and classified. In order to correct theamplitude and phase error between two orthogonal channels such as horizontal and verticalchannels, an iterative algorithm implemented in time-frequency domain is brought in to getthe balance between the two orthogonal channels. After SAR imaging, the measuredscattering matrix is usually dissymmetrical, and the Cameron decomposition method can dealwith this problem as an optimization correction method for scattering reciprocity. Utilizing theCameron decomposition, an initial classification result confirming to the inherent physicalpolarimetric characteristics of the targets is produced. According to the statistic feature, themaximum-likelihood (ML) classifier under the complex Gaussian distribution is employed toclassify the pixels in PolSAR image based on the initial classification result. The feasibility ofthe approach is demonstrated with experiments.Under the assumption of a Wishart distribution of a polarimetric covariance matrix, it iscan be found empirically that ocean’s span histogram can be well modeled with theexponential distribution. The probability density function (PDF) of the exponentialdistribution usually has a peak at zero with a decreasing tail, which exhibits a sparse feature.Since the eigenvalue of the covariance matrix is consistent with the span, its histogramdistribution also can be regarded as a sparse distribution. From the histogram, the sparsedegree can be estimated. Combing with the NMF process, the main feature of the PolSARocean data can be extracted for ship detection. Both quad polarimetric and dual polarimetricocean data sets are experimented to prove the efficiency of the proposed approach.A study focusing on the application of H/α decomposition to the original compactpolarimetric (CP) SAR data is studied. CP mode has advantages over a fully polarimetricmode in terms of reductions of pulse repetition frequency, data volume, and system powerneeds, meanwhile, fully polarimetric information can be approximately reconstructed fromthe CP mode. However, two assumptions should be followed during the reconstructionprocess, which leads to the loss of the information and increases the computational burden.Hence, the H/α plane is modified and re-plotted out to distinguish different scatteringmechanisms using the original CP data. The feasibility and effectiveness of the new H/αdecomposition for CP mode are analyzed and verified by experiments with measured data.