Research On Polsar Image Classification Based On Target Decomposition

Polarimetric synthetic aperture radar(Pol SAR) with its advantages of all-weather, all-time has become most advanced technology in the field of remote sensing. It can achieve polarization information through receiving different combinations of electromagnetic waves. Also, it records the reflected echo’s amplitude and phase. Because it can achieve polarization information, the targets is more fully described. Pol SAR has been successfully applied to the many fields, such as agriculture, city planning, and even military field. With the rapid development of polarimetric synthetic aperture radar’s hardware, the ability which is used to interpret radar data is obviously not enough. In terrain classification field, we are facing many problems, such as the feature can’t fully describe the target, classification accuracy and performance is not high enough, the robustness of the algorithm is not strong enough and so on. Based on the theory of the decomposition of target, we studied classification of polarimetric synthetic aperture radar images, and the main contents are:1. Coherent and incoherent decomposition theory has been studied and analyzed through experiments. We studied a super-resolution decomposition algorithm which is using space correlation and Pauli decomposition. It can classify different scattering mechanisms which are mixed in one cell. We successfully use this idea based on non-coherent decomposition, the Freeman decomposition. Its performance can be proved though related experiments.2. We deeply studied some methods which are based on eigenvalue decomposition, obtained three parameters, polarimetric scattering entropy H, average scattering angle (?) and the degree of anisotropy parameters A. Use those three parameters, the classification plane and three-dimensional classification space has been constituted. This paper also discusses the influence of speckle noise above three parameters. After the introduction of Wishart classifier, we constituted two unsupervised target decomposition scheme combined three parameters and coherent and non-coherent decomposition theory. Finally, the advantages and disadvantages of the related parameters were studied, and the corresponding parameter optimization solutions were given.3. The support vector machine classifier is introduced into the polarization Synthetic Aperture Radar feature category, which is due to the classifier has many characteristics, such as sparse samples, high-latitude. Due to target decomposition theory can provide good classifier feature, we combine the target decomposition theory and support vector machine together. A lot of experiments had been done based on these and the results are good. We used genetic algorithm to optimize some parameters in support vector machine. Finally, a novel unsupervised classification method with good classification performance was introduced, this method estimate features’ quality using power and Wishart cluster centers.