Classification Of Polsar Image Based On Target Decomposition、object-oriented And Decision Tree Algorithm

The information of Polarization Synthetic Aperture Radar （PolSAR） data is muchgreater than that of SAR data, and the existing classification method partly use theinformation hidden in PolSAR data. To fully utilize the information in the PolSAR data,this thesis proposed a new PolSAR data classification method which combines targetdecomposition theory, object-oriented methodology and C5.0decision tree algorithm.First of all to the full polarization image according to different polarizationdecomposition manner to decomposition and extract all kinds of polarization parameterscontain features scattering mechanism information; Then making a segmentation to thepolarization parameters images, which makes image processing unit from pixel to object;Finally using C5.0decision tree algorithm to establish decision rule and then realizeterrain classification This paper compares the proposed method with the classical H/αunsupervised classification、supervised classification and the eCognition classificationusing RADARSAT-2full polarization data. The results show that the proposed methodreached83.2%, higher than the other three classification8.59%,7.67%and1.29%respectively. The experiment shows that the proposed method has higher precision andstronger applicability. This method makes contribution to the updating of landmonitoring, land resources investigation and fundamental geographic information.