Independent And Sufficient Target Decomposition Of PolSAR Image And Its Applications

Polarimetric synthetic aperture radar(Pol SAR) is developing rapidly over the recent years, and has been one of hot topics in remote sensing area. Polarimetric features extracted using polarimetric target decomposition are the basis of Pol SAR target detection and image classification. How to extract effective polarimetric features is the key to follow-up image interpretation. With the development of Pol SAR system, Pol SAR image resolution is also improving. In order to extract features of high-resolution Pol SAR image accurately, a coherent target decomposition method which can extract independent, sufficient and meaningful features is needed. At the same time, how to apply these features to image interpretation has yet to be a breakthrough. Therefore, it is important to carry out independent and sufficient target decomposition method and apply it to image interpretation. This thesis is based on an independent and sufficient target decomposition(ISTD), focusing on target decomposition of rotated building area, image classification and man-made target detection.Traditional coherent target decomposition methods suffer from dependent parameter problem and insufficient parameter problem, meanwhile, complex representation of traditional coherent target decomposition methods is hard to understand. In order to overcome these problems, a mapping from complex filed to real filed which is derived by group theory is used. Based on the mapping, representation of typical targets can be easy to understand. After that, an ISTD method is proposed, the parameters decomposed using the ISTD method are proved to be independent. Based on ISTD, an incoherent ISTD(IC-ISTD) is proposed using eigen-decomposition. Pol SAR images are used to compare the results of ISTD and traditional target decomposition methods. Experimental results show the validation of the proposed methods. The results of the proposed target decomposition methods can be used as polarimetric features of further image interpretation.Overestimation of volume power on rotated building area by model-based target decomposition is researched. The cause of the overestimation problem is analyzed. Based on this and the fact that some parameters of IC-ISTD results are independent with orientation angle, a mapping linking model-based decomposition method and IC-ISTD results is proposed to solve influence of rotation angle on model-based target decomposition. The mapping turns model inversion problem to be a mapping problem, then an IC-ISTD based fourcomponent decomposition method is proposed using this mapping. The proposed method is effective against highly rotated building area. Pol SAR images are applied to compare the IC-ISTD based four-component decomposition method and traditional model-based target decomposition methods. Experimental results show that volume power in rotated building area is suppressed by using IC-ISTD based four-component decomposition, therefore, building can be distinguished from forest easily.ISTD is applied on image interpretation. In classification application, features extracted using traditional feature extraction methods suffer from information loss or information redundant problem. Based on the analyses, an ICISTD based feature set is construct using IC-ISTD. The features can be used to fully interpret target-related feature. After that, a feature weighted random forest classification method is proposed to apply IC-ISTD based features on classification application. Pol SAR images are used to compare proposed classification method and traditional classification methods. Experimental results show that, compared with traditional classification methods, the proposed feature weighted random forest classification has better performance on classification accuracy. Man-made target scattering properties and orientation angle of different kinds of targets are analyzed. An image segmentation method based on the results of target decomposition is proposed. After that, a rotation angle based man-made target automatic detection method is proposed by using the value of rotation angle and the homogeneity of rotation angle. Experimental results show that the proposed man-made target decomposition method can effectively remove false alarm by branch-trunk structure and improve detection rate of rotated building area.