POLSAR Image Classification Based On Complete Decomposition Of Coherent Matrix Model

As a kind of advanced microwave remote sensing system,polarimetric synthetic aperture radar(POLSAR)can obtain complete electromagnetic scattering characteristics of target and high resolution image at the same time.Based on incoherent scattering model decomposition is an important means of polarization information extracted from PolSAR image.By getting different scattering mechanism of power and other parameters to class the PolSAR image.This paper mainly studies based on incoherent scattering model decomposition and image classification method.At present,based on incoherent scattering model decomposition have two questions:(1)Because it is under the condition of the reflection symmetry polarization decomposition,regardless of the cross polarization information,can cause volume scattering power value is too high,so difficult to distinguish between forest area,and PolSAR with the different direction of the scattering characteristics of buildings,etc.(2)Due to not following the nonnegative eigenvalue decomposition,can make the scattering mechanism has a negative power,it can also cause volume scattered power value estimate is too high,reduce the availability of decomposition results.Aiming at these problems,according to the coherent scattering matrix decomposition theory,a newl method was put forward,In view of the two parts,the first part is under the reflection symmetry polarization decomposition,consider the cross polarization phase information,the correlation matrix of nine parameters all scattered power calculation.The second part is the nonnegative eigenvalue constraint scattering parameters on the calculation of reference to the volume scattering,choose to make the minimum residual matrix is positive semi-definite matrix eigenvalues for scattered power,ensure that three conditions of nonnegative eigenvalue of scattering mechanism.Through each pixels,the scattering matrix calculation of directional Angle,directional and decide to cross polarization phase after the positive and negative,to further distinguish the surface scattering and double scattering.Using Radarsat-2 PolSAR data of experimental verification,compared to the decomposition of Freeman,Yamaguchi decomposition,Van Zyl decomposition,as a result,this method can make the city area double scattering power value increased significantly,inhibition of scattered power to obtain the very good forest part.Based on the above decomposition algorithm to get the three components,used in PolSAR image classification.As part of the target object is the performance of mix scattering mechanism,so the classification not only will dominate the scattering mechanism of Max(Pv,Pd,Ps)as the target scattering mechanism.So,this paper presents two improved methods.The first method is to dominate the scattering mechanism and secondary scattering mechanism of classification method of plane target feature is divided into nine different types of scattering,and Wishart classifier iteration fine classification.The second method is,considering scattering entropy H is reflect the randomness of target object,categorizing entropy H and scattering mechanism of plane,and the fine Wishart classification.Respectively in this paper,the decomposition algorithm and Freeman decomposition of three-component for P?/P? Wishart andH/P? Wishart experiment.For town area of experiments show that the decomposition algorithm of the classification accuracy is higher than the classification accuracy of Freeman decomposition,P?/P? Wishart overall classification accuracy is higher than H/P? Wishart classification accuracy.Combination of completely coherent matrix decomposition P?/P? Wishart classification accuracy,best for scattering mechanism similar to that of road and water body are correct,the way of vegetation and vegetation area effectively distinguish to correct for toward different buildings.