POLSAR Images Classification Based On Feature Statistic

Polarimetrie Synthetic Aperture Radar(POLSAR)images describe an interrelation between electromagnetic waves launched by SAR system and terrains. It mainly presents a distribution characteristic of terrains. So, we can classify and recognize terrain targets by studying POLSAR images. However, many complex scenes, which is caused by the interference among electromagnetic waves and the clutter distribution of terrains, exist in POL-SAR images. The scenes will not only reduce the classification and recognition accuracy of POLSAR images but also hinder the classification technology development of POLSAR images. Thus, in this paper, we propose three methods to classify the complex scenes in POLSAR images. The main content is introduced as follow:(1) We propose a new method based on the joint statistical properties of eigenvalues to classify the similar scenes in POLSAR images. This method estimate the distribution of volume scattering power and maximum eigenvalue based on the joint Gaussian model. Under the statistical framework, we use Bayesian classifier to classifier the similar scenes in POLSAR images. In order to make full use of the local texture information, we propose a local statistical method based on MRF model to improve the classification accuracy of POLSAR images and regional consistency.(2) We propose a classification approach on the heterogeneous scenes of POLSAR images by using the local Gaussian model. Firstly, this algorithm measures the space heterogeneity of POLSAR images based on the initial classification result by using the local statistical method. Secondly, we set an adaptive threshold to classify the heterogeneous scenes in POLSAR images. The contribution of our method is that it can describe effectively the variety and volatility of the heterogeneous scenes by analyzing the distribution density of the main terrain in local space.(3) We propose a classification and recognition method for the heterogeneous scenes in POLSAR images based on the evidence theory. This method uses the uncertainty theory to measure the heterogeneity of POLSAR data. On the basis of evidence theory, the belief function expressed by Euclidean distance of the maximum eigenvalue is used to identify the attribute of the object pixels. Finally, according to the maximum belief rule, we classify and recognize the heterogeneity scenes of POLSAR images. The basis theory of this method is solid. Our method not only can classify the heterogeneous scenes and can recognize its terrains.