Skip to Main Content
This paper develops a classification way using polarimetric synthetic aperture radar (SAR) data. Polarization of radar electromagnetic wave is a very important factor of backscattering theory and geosciences applications. However, polarimetric SAR images are still difficult to interpret and therefore there is a need to improve the classifier that can make use of the polarimetric information. Under the certain circumstance the present paper attempts to classify remotely sensed scenes by all the complete polarization response parameters, which are presented as data dimensions for classification arithmetic. The test site is located in Hetian of Xinjiang, China. SIR-C data were acquired in the test site in 1994. Fully polarimetric SAR data can be processed as multi-dimensions images. But the polarimetric information is related greatly with variable targets. Firstly, we decompose the backscattering matrix and get polarimetric ratios, polarimetric degree, and polarimetric entropy. These polarimetric parameters are considered as data dimensions in which elements change in terms of probability functions with variable targets. Training samples were then generated from the outputs of the unsupervised classification of K-means, to be used in subsequent supervised classifications of two frequencies (C- and L-band for SIR-C data) and various polarization combinations. The estimation of the classifier tallies with the local municipal statistics. The result shows that classification precision can be improved finely with the polarimetric technique.