Research On Classification Method Of Fully Polarimetric SAR Images

The use of Polarimetric SAR(PolSAR)images for land classification is one of the most important applications of remote sensing,and it is also a very active research area.Compared with ordinary remote sensing images,the active side-viewing SAR is extremely sensitive to the dielectric and geometric characteristics of the land surface structure and roughness,so the backscattering and polarization information of the SAR data can be used to provide additional information for classification of land objects.Has research value.PolSAR can distinguish the scattering mechanism of different ground types,which is more suitable for LULC classification than traditional SAR and single-polarization SAR.This paper takes the northwestern part of Pengzhou City of Sichuan Province as the research area,and takes the ALOS-PALSAR polarimetric image as the data source.Based on the systematic analysis of domestic and foreign research status,the polarization SAR decomposition theory and the classification of land objects are studied.Made some constructive suggestions and improvement methods.(1)According to the characteristics of polarimetric SAR data,ALOS-PALSAR data is classified by H/? using classical classification methods,H/?-Wishart classification,H/?/A-Wishart classification,and analysis of all kinds of features in the study area.The scattering characteristics and their distributions were compared and the results of classification and H/? plane scatter plots were compared and analyzed.(2)The traditional BP neural network classification method is improved,and four parameters ?1,?2,?3 and SPAN are selected as training samples according to Cloude decomposition,and the improved L-M algorithm is used to improve the convergence speed for classification by BP neural network,and the conventional solution is solved.The problem of slow convergence rate and poor classification effect of BP neural network classifiers.(3)In order to make full use of the polarization characteristics,texture features,polarization parameters and other information of polarimetric SAR images,a method based on multi-feature combination and SVM is proposed to classify polarimetric SAR image features.(4)At the same time,the scattering characteristics of the mountain shadow and the water body are similarly caused for the SAR imaging system,and confusion occurs in the classification process.A method of using the DEM terrain modeling and introducing an improved Otsu algorithm to segment the DEM is designed to polarimetric SAR image.The classification of land objects was studied.The experimental results show that,for the classification of polarographic imagery in mountainous areas,the discrimination degree of obscure ground objects in the classification can be effectively improved,and the classification effect and classification accuracy of mountainous land features can be improved.