Choosing Yutian Oasis as the study area, the polarimetric characteristic parameters are obtained from quad-polarized PALSAR-2 data in a variety of target polarization decomposition treatments. 11 polarized parameters with lesser noise are selected as the best classification features through visual interpretation to train the support vector machine classification. Wishart classification, support vector machine classification and several polarimetric decomposition methods are combined to extract the different levels of soil salinization information. Classification results are quantitatively analyzed and validated by Landsat-8 operational land imager image combined with visual interpretation and field investigation. The analysis results of confusion matrix show that, comparing with Wishart classification, the support vector machine classification increases the classification accuracy and Kappa coefficient from 80.48% to 88.00% and 0.73 to 0.83 respectively. It is illustrated that the individual coherent decomposition cannot fully exploit the rich information of PALSAR-2 data, and the good classification results can be achieved by using the target polarimetric decomposition parameter for features classification process. Using fully polarimetric PALSAR-2 data, the method combining target polarimetric decomposition with support vector machine classification has advantage to extracting salinization information.