Abstract: Two classification models（namely Krogager-SVM and Pauli-SVM） on bases of Support Vector Machine （SVM） wereproposed and conductedrespectively,through using Krogager and Pauli polarization decomposition methods. A fully polarimetric synthetic aperture radar（SAR） remote sensing data was utilized over the study area（on the delta oasis between the Weigan and Kuche River in Xinjiang, China）,and different degrees of salinizized soil information in the typical oasis of arid area was extacted. Then by adopting the field verification data comparison and corresponding analysis was conducted between the classification results of proposed methodology and traditional SVM classification method. The results show that the Krogager-SVM and Pauli-SVM classification models improved classification accuracy in contrast with the traditional classification method for soil salinization extraction in the arid regions,and the overall accuracy enhanced from 74.17% to 80.598 0% and 82.387 6% respectively（increased by 6.43% and 8.21%,and the kappa coefficients increased by 0.08 and 0.12 respectively）. This indicates that the classification models proposed in this paper have some potential in the soil salinization extraction by using fully PolSAR data.

Key words: salinization, PolSAR, Pauli decomposition, Krogagger decomposition, SVM