Polarimetric SAR Image Classification Based On Freeman Decomposition

Polarimetric Synthetic Aperture Radar (POLSAR) is a muti-parameter, muti-channel imaging radar system, which obtain the polarization information of the target by measuring the full polarization scattering echo of each ground resolution cell, such as scattering matrix, polarization coherent matrix or kelmaugh matrix and so on. The obtained polarimetric SAR image data can provide more surface feature information than conventional radar images, so errain classification plays an important role in POLSAR data application.This paper mainly study SAR image clasification methods based on the combination of statistics characteristics and physical scattering characteristics. Taking into account the cluster analysis methods and the polarization scattering mechanism, this paper proposed several improved classification methods based on the Freeman decomposition, which mainly includes the following three aspects:1. We proposed a method based on the Freeman-durden decomposition and adaptive neighborhood fuzzy MRF, which achieve an effective combination of polarization decomposition and cluster analysis methods, and use the three scattering powers as the classification features of fuzzy MRF. Apply FCM to initial classify the image to avoid the impact of noise when direct use of MRF, and then use the Fuzzy Markov Random Field (FMRF) clustering method to obtain the image classification. In addition, we introduce the adaptive neighborhood into Fuzzy Markov Random Field to further improve the classification result. This method has the advantage of high adaptability and good anti-noise performance.2. We proposed a method based on the Freeman-durden decomposition and the co-polarization ratio. Apply the three scattering powers to achieve the initial classification at first. Then use the co-polarization ratio to classify the image into more classes. Finally, to improve the classification accuracy, the data sets of which are classified by an iterative algorithm based on a complex wishart density function. Compared with the classification results of other traditional methods, the proposed algorithm ideal is easy to understand and has better classification accurancy.3. We proposed a method based on the scattering power entropy and the co-polarization ratio, which introduce the scattering power entropy in initial classification, so it can classify the maxied scattering mechansim pixels more effectively. Combined with the co-polarization ratio, a better initial segmentation can be achieved. In addition, an improved class number reduction technique is applied on the initial resulting clusters to obtain an appropriate number of categories. Three real polarimetric SAR images are tested, the experimental results show that high precision, high adaptability and better connectivity are observed of the proposed method.This work was supported in part by the National Natural Science Foundation of China under grants61003198,61173092; the Fundamental Research Funds for the Central Universities JY10000902045; The Fund for Foreign Scholars in University Research and Teaching Programs under grant B07048.