Classification Capability Analysis of Polarimetric Features Obtained by Decision Tree

Abstract

The decision tree based on polarimetric features achieves hierarchical classification of vegetation by utilizing the different scattering mechanism. Because the observation capability of the certain POLSAR (Polarimetric Synthetic Aperture Radar) sensor keeps stable, the polarimetric features for classification can probably be applied to different data using the same sensor. The transplantation of the polarimetric features in the tree nodes not only achieves high classification accuracy, but also saves training time. In this paper, two sets of AIRSAR data in Flevoland area are used to verify the classification capability of polarimetric features obtained by decision tree nodes. The training samples of the first data set is used to train a two-dimensional decision tree with two-dimensional polarimetric features at the tree nodes, which could certainly achieve good classification accuracy for the testing samples in a supervised way. Then the features in this constructed tree nodes are employed directly to classify those vegetation types in the second data set without any training process. The experimental results show that the classification rules and polarimetric features used to separate the same two vegetation types are applicable to another data set, which is of great significance to the study of the role of polarization features in classification.