PolSAR Image Classification Based On Tensor Decomposition

Polarimetric synthetic aperture radar(POLSAR)transmit and receive radar signals in different polarization modes, so that the radar system can get more information about the target. The polarimetric information can be recorded in the form of the complex scattering matrices, with which the scattering characteristic of the targets can be extracted and analyzed. Because of providing the whole scattering characteristic of the targets, the polarimetric SAR is of great importance in the remote sensing, which can be used widely in both civil and military fields. When using traditional image processing method, the original data is converted into vector, but the relationship between the adjacent position data in the original data will be destroyed, and loss of the space structure information in the original data space. In order to sove the question, the tensor model is introduced into the classification of polarimetric SAR images. Because the tensor has an advantage in dealing with high dimensional data,it will improve the performance of the algorithm. The primary contributions of the dissertation are as follows:(1)We propse a new based on tensor decomposition to classify the POLSAR images.different with existing methods,we need not transform the original data into vector,use tonser model expressed the original data, to reserve geometric space information of three channels of data from the Polarimetric SAR data. The data of three channels of polarization SAR image are fully utilized to improve resistance noise and reduce the number of training samples and improve the accuracy of classification.(2) We propose a classification method for POLSAR images based on non negative tensor decomposition. The non negative tensor decomposition of the tensor data into the non negative matrix and the non negative weight. These non negative base tensor is sparse, and the non negative tensor decomposition has the uniqueness result. We Using non-negative tensor decomposition algorithm to extract features,then initial classify the image,Finally,to improve the classification accuracy,the data sets of which are classified by an iterative algorithm based on a complex wishart density function.