| Polarimetric Synthetic Aperture Radar(PolSAR)imaging is not affected by climate,and has high image resolution.Therefore,it has a wide range of applications in disaster monitoring and resource exploration and so on.Polarimetric SAR can measure the scattering information of each resolution unit under four polarization states.It is an important part of interpreting polarimetric SAR image to classify the objects in polarimetric SAR image according to scattering information.The purpose of this paper is to classify polarimetric SAR image.This paper focuses on the feature extraction and classification methods of polarimetric SAR image.The main work and contributions are listed as follows:1.The theories of electromagnetic waves related to polarimetric SAR are introduced,including the representation of electromagnetic wave,the relationship between incident wave and echo,and the typical scattering mechanism.2.The high dimension features of the polarimetric SAR image are extracted.This paper analyzes the various polarimetric target decomposition theorems,including Huynen decomposition,Freeman decomposition,Cloud decomposition and Pauli decomposition,etc.A single decomposition theory can only obtain a few specific polarimetric information.Therefore,this paper combines multiple decomposition scattering mechanisms to obtain part of the features for the polarimetric SAR image.The rest part of the features consisted of the deformation of each element in the scattering matrix,including total scattering power,amplitude,phase,polarization ratio,etc.3.A new dimension reduction algorithm is proposed for the high dimension features.The paper analyzes the shortcomings of the LLE dimension reduction algorithm and proposes the FWLLE dimension reduction algorithm.Based on Freeman decomposition and Wishart classifier,the algorithm clusters data into multiple clusters and improves the LLE algorithm by changing the distance metric of LLE algorithm.Finally,it is analyzed that the data after dimension reduction using FWLLE algorithm is more suitable for polarimetric SAR image classification.4.Compared with the pixel-level classification methods,combined with the spatial information of the target in the image,the region-based classification methods have obvious advantages.This paper introduces the Majority Vote(SLIC-MV)method and proposes two classification methods,including Bag of Words model method(SLIC-BoW),the method of combining Majority Vote and Wishart classifier(SLIC-WMV).These three methods are based on superpixel segmentation,and the final classification unit is the superpixel block.Through the analyses of experiment results,the proposed two classification methods can effectively overcome the influence of speckle noise.Their classification accuracy is higher than the pixel-level classification methods and the SLIC-MV classification method,especially the SLIC-WMV classification method has the best classification result. |
