The decision tree model has great significance in the application of polarimetric SAR data classification, whose results in many types of classification applications obtain good accuracy and are interpretable by polarimetric scattering mechanisms. In the traditional decision tree model, because one single feature is employed by the nodes of the decision tree, the accuracy of the classification result tends to be poor, especially, for applications that classify objects with similar scattering characteristics. In this paper, we propose an improved method to create a two-dimensional vector of features instead of one single feature at the decision nodes. As a result, the classification results of the new method adopting the same feature set as the traditional decision tree can achieve better accuracy. In addition, after classification, the new method may employ a confusion matrix to identify the decision node that yields a classification error, which will facilitate the objectoriented feedback adjustment of classification results, thus making it possible to improve the classification accuracy of the specified object. Our experimental results with AIRSAR-Flevoland data prove the validity of the proposed method, and we draw some useful conclusions about the scattering characteristics of several types of vegetation.