Classification Of PolSAR Images Based On Sparse Representations And Contextual Information

Polarimetric synthetic aperture radar(Pol SAR) has great ability to get comprehensive geographic information by the SAR images getting from the four different polarimetric channels. Therefore, it has a great prospect in the application of the research about seismology, military, agriculture, forestry, hydrology etc. The classification of Pol SAR image is the precondition and basis of the following Pol SAR image interpretation. Presently, there are still some problems about the classification of Pol SAR image such as the problem of classification of high dimensional and nonlinear data. Recently, sparse representation has become a highly effective and innovative tool in the field of pattern recognition. The purpose of this paper is to promote the application and development of Pol SAR image interpretation with the features that extracted with covariance matrix and coherent matrix and the characteristics of the sparse representation.This paper mainly includes the following three aspects:Firstly, the method of the expression of Pol SAR image is studied which includes scattering matrixes and the second-order statistical matrix from the basic theory of the Pol SAR technique. Then the feature extraction algorithm including polarimetric features and non-polarimetric features. The polarimetric features are extracted by the target decomposition methods based on the scattering models and the eigenvalue decomposition. The non-polarimetric features are extracted by the Gray Level Co-occurrence Matrix(GLCM) which can obtain the textural features.Then the basic theory of sparse representation is studie d from the aspects of the sparse representation of signals, the design of the dictionary, the algorithm to obtain the sparse coefficients and the classification model based on sparse representations. And the Polarimetric SAR image classification model base d on sparse representation is proposed in this section with the combination of the Polarimetric SAR features. The data of EMISAR in Foulum, AIRSAR in San Francisco and Flevoland are used to validate the algorithm.Finally, the sparse constrained optimization model-Laplace model and contextual sparse representation are studied for promoting the efficiency and effectiveness of sparse representation classification model. And the Polarimetric SAR image classification model based on contextual sparse representat ion is proposed. In this paper, the data of EMISAR in Foulum, AIRSAR in San Francisco and Flevoland are used to verify the above theory and the experiments show that the proposed algorithms for Pol SAR image classification are useful and effective.