Research On PolSAR Image Features Analysis And Earth's Surface Targets Classification

Fully Polarimetric Synthetic Aperture Radar (SAR), a kind of advanced Earth observation instrument, can obtain scattering characteristics of each resolution cell under any polarization state. And its remote sensing techniques offer efficient and reliable means of collecting information required to extract biophysical and geophysical feature parameters about the earth's surface. This dissertation discusses different ways of extracting Polarimetric scattering features, analyzes the inner relation between targets'classes and its scattering features, extractes kernel method mapped features and classifys earth's surface in the condition of no changes on the target's scattering characteristics. Meanwhile, the tool of formal concept analysis is introduced and employed to mine classification rules of object recognition based on the association between scattering features and targets'classes. And the method of mining classification rules provides a new idea and direction at Polarimetric scattering mechanisms further study, feature subsets optimization, targets classification and object recognition.The main work is listed as follows:1. Introduce the fundamental theory of electromagnetic wave polarizer, including the description of the EM wave's polarization states, different kinds of the typical representation of Polarimetric SAR data and some elementary targets presenting canonical scattering mechanisms. And then, analyze the statistical properties of Polarimetric SAR images and validate the feasibility about using the statistical characteristics calculated directly from the Polarimetric coherency matrices (or covariance matrices) to classify and identify scattering targets.2. Analyze two main kinds of Polarimetric target decomposition theorems. One is the important class of target decomposition theorems based on Eigenvalues of the coherency matrix T3 (Cloude and Pottier), and the other one is based on a"model-based"decomposition of the covariance matrix C3 or the coherency matrix T3 (Freeman and Durden). And then analyze those target decomposition methods's principles, calculation process and the importance in targets classification.3. Employ generalized discriminated analysisto extract Polarimetric characteristics. This new feature extraction method uses a kernel function mapping the coherency matrix to high-dimensional space to extract new Polarimetric features. Based on this, a new classification is proposed, which maintains Polarimetric scattering characteristics on the consideration of targets' physical scattering mechanisms. 4. Formal concept analysis is employed in PolSAR data mining and rules extraction. After analyzing the link between Polarimetric characteristics information and the surface targets, this tool constructs the concept lattice of the Polarimetric classification formal content. And then, apply this mined classification rules in surface targets classification research. In addition, this thesis initially explores the application of classification rules in classification accuracy assessment, optimization features subset.