Object-Oriented Classification of Polarimetric SAR Imagery Based on Texture Features

This paper presents a new object-oriented classification of Polarimetric Synthetic Aperture Radar (PolSAR) data method based on region-merging technique for multiresolution segmentation and Decision Tree Classifier (DTC) for classification. This new approach can overcome some limitations of traditional pixel-based classification for high resolution PolSAR data. This research takes the test image of an area in Hainan, which resolution is better than 1 meter, acquired by the first airborn PolSAR sensor of our country. The data processing has four steps. Firstly, establishes the classification hierarchy..In this step, it takes the region-merging technique for multiresolution segmentation and mean-variance method for determining the optimal scale of every class. Secondly, extracts polarimetric features, texture features and all kinds of statistical features of objects. Finally, classifies by DTC, which inputs is some features selected from above features. In this test, the overall accuracy is 92.8%, which shows that the method proposed in this paper can improve the classification accuracy to some extent.