Fully Polarimetric SAR Image Classification Based On Joint Sparse

Polarization synthetic Aperture Radar(POLSAR)is an advanced means to obtain remote sensing information and work all day,which is not affected by the weather.The system of the polarization synthetic Aperture Radar has four different ways to send and receive,they are HH,HV,VV,VH,the fully polarization SAR can record the scattering information of the objects in these four polarization combinations,and the target backward scattering record completely,because HV and VH are reciprocal,how to study the relationship under the three HH,HV,VV different combination,use joint sparse on the three channels' similar structure information and combined with the polarization characteristics,several improved methods of full polarized SAR classification are proposed and applied to coastal zone SAR data and multi-temporal full polarization SAR data,which mainly include the following:Frist,proposed a method of coastal zone classification of full polarization SAR data based on three channels joint sparse,mainly introduces the joint sparse model of three channels and the joint sparse model based on the polarization characteristics.This chapter mainly use some terrains of coastal zone which have strong texture features but similarity weak scattering characteristics to construct joint sparse model on three channels,and combined with the joint sparse model based on the polarization characteristics.This method has better classification accuracy on these coastal zone types which have larger economic interests.Second,proposed a nearest regularized joint sparse representation on multi-temporal full polarization SAR based on dictionary transfer.We put forward a nearest regularized joint sparse representation model on the basis of the joint sparse,learning a dictionary on the source domain of the multi-temporal SAR image migrate to the target domain of multi-temporal SAR image,at the same time update the dictionary for the classification of the target domain multi-temporal SAR image.The nearest regularized joint sparse representation on the three channels can reduce the error brought by the difference between three channels compared to regular joint sparse,and this method has a certain advantage in classification compared to the classical methods,and it can save time to resample training according to the dictionary transfer.Third,proposed a multi-temporal full polarization SAR change detection and target recognition method based on classification,put forward a target recognition model based on sparse representation,mainly using the minimum sparse reconstruction residuals and the average residual of every type to be the constraint,to deal with the difference of polarization SAR image for showing the changed parts.Traditional change detection is only to find out difference and show it,this work will not only detect the changed area,but also use sparse representation to recognize target of the changed area based on an effective dictionary,so it can get detailed division of polarization SAR,the method is easy to implement,think simply,and understand easily.