Synthesis of multi-source remote sensing data for classification based on Bayesian theory and MRF
结合贝叶斯理论和MRF的主被动遥感数据协同分类

The iterative technique for multi-source remote sensing data classification is presented in accordance with the advantages of multi-source data in feature extraction. In the method, the Advanced Synthetic Aperture Radar (ASAR) backscatter coefficient is normalized by the incident angle at first. Then, a classifier based on the Bayesian theory and Markov random fields (MRF) is developed, and the Vertical-Vertical, Vertical-Horizontal (VV, VH) polarizations of ASAR and all the seven TM bands are used as inputs of the classifier to get the class labels of each pixel of the images. Finally, the method is validate, the necessities of normalization and integration of TM and ASAR are discussed. The results show that the precision of classification in this paper is 89.4%, which is increased by 4.1% and 11.5% compared with the methods of without normalization and using single TM data. These analyses illustrate that synthesis of multi-souce remote sensing data is an efficient classification method.