The PolSAR Object-oriented Classification Of Homogenous Area Segmentation Research

High-resolution Full-polarization Synthetic Aperture Radar (PolSAR) feature has richer image details, and there are a lot of speckle noise, The traditional pixel-based processing technology can not effectively suppress speckle noise, what’s more, segmentation has a lot of disadvantage,such as over segmentation or lacking segmentation, and result of this disadvantage is the wrong classification when we classify the image. In order to improve the segmentation and classification of PolSAR images, this paper proposed a object-oriented analysis technology, which can effectively solve the problems faced in the process of pixel-based processing. The object-oriented analysis technique uses pixel set(homogeneous area) as the unit of analysis. It plays a key role in addressing the high-resolution image PolSAR scale effect and suppressing speckle noise. There are some progress in research the PolSAR object-oriented analysis technology, but the technology for image PolSAR application is not yet mature, still requires further research and improvement.First, the characteristics of PolSAR data are analyzed in the paper, and then we briefly introduced the object oriented analysis technology, still then we analyzed and filtered the optimal feature subset for the object-oriented classification of PoLSAR images. Strategy analysis is that 12 groups feature subset that is consists of a group of single and combined features are bulit, we trend to use the Mahalanobis distance classifier to classify the 12 sets of feature subset. The results show that we can extract the optimum feature subset. Because the fuzzy decision algorithm can make full use of the variety of characteristic information of PolSAR images, this study choose a fuzzy decision algorithm for the PolSAR image segmentation of homogeneous region based on the selected subset. Finally, compared with other common segmentation algorithms, the effectiveness of the proposed method is verified. Based on the optimal feature subset of PolSAR images, We are sure that the homogeneous region segmentation in a great extent can have good effect for subsequent targets classification or recognition.Of course, this experiment also has some limitations, so we need more experimental data to verify the effectiveness and versatility of this method in the future.