Polarimetric InSAR Land Type Classification And Interpretation Based On Feature Library

As one of the most advanced sensors in the field of remote sensing,Polarimetric Interferometric Synthetic Aperture Radar(Pol In SAR)possesses both the advantages of Interferometric Synthetic Aperture Radar(In SAR)and Polarimetric Synthetic Aperture Radar(Pol SAR)and is extremely sensitive to the ground characteristics,providing it great significance and broad prospect both for civil use and military use.Pol In SAR land type classification,as the first step of Pol In SAR information processing and one of the most important applications in remote sensing as well,is now one of the most active areas in Pol In SAR.Pol In SAR land type classification can be classified into supervised classification and unsupervised classification.Due to the land type recognition problem,both methods involve lots of experts in Pol In SAR image recognition,which greatly limits the Pol In SAR land type classification application.The use of prior information solves the problem dramatically.The prior information is either labeled observation data or scattering models,which can be regarded as the features of the data.Due to the lack of uniform framework,the features cannot always be collected and employed fully and effectively.To take the advantage of the prior information systematically,the features set is regarded as feature library,and how to build and use it is deeply researched into.In the framework of feature library,a classification method based on feature library is proposed.The model-based polarimetric decomposition is further researched into because the performance of the classification method relies heavily on it.A decomposition flow for the feature library and a refined model for vegetation area are proposed.The main content and accomplishment of the dissertation are as follows:1)The feature library is defined as the set of features carrying prior information,which is researched into independently on how to build and use it fully and systematically.The research boosts the development of the classification method and scattering model as well.The features in the feature library consist of labeled observation data and scattering models,and how to build feature library based on the two is researched into.A feature library building method combining the two is then proposed.2)In the framework of feature library,a complete classification method is proposed combining the ideas of classification based on scattering feature,classification based on statistics,classification based on superpixel and staged classification,the advantage of which is verified by comparing the results to that of the supervised classification.A decomposition flow is further proposed based on the above classification method,solving the common problems of model-based decomposition.3)In the framework of feature library,the scattering models of non-vegetated area and vegetated area are summarized and their advantages and disadvantages when dealing with refined classification problem are presented.How to improve the model is proposed based on the discussion above.An X-Band vegetated area refined scattering model is then proposed by realizing the improvement,the advantage of which is verified by comparing the result to that of the Freeman-Durden decomposition.