基于目标分解理论的全极化SAR图像神经网络分类方法

由于全极化合成孔径雷达(synthetic aperture radar)能够测量每一观测目标的全散射矩阵，即可合成包括线性极化、圆极化及椭圆极化在内的多种极化图像，因此与常规的单极化和多极化SAR相比，在雷达目标探测、识别，纹理特征和几何参数的提取等方面，全极化SAR均具有很多优点，但是由于地物分布的复杂性往往造成不同地物具有相似的后向散射信号特征，因而加大了地物信息提取的难度。同时由于这些极化合成图像具有较高的相关性，从而导致了图像分类精度的降低。为了提高全极化SAR图像的分类精度，基于新疆和田地区的SIR-CL波段全极化雷达数据，利用目标分解理论首先将地物回波的复杂散射过程分解为几种互不相关的单一的散射分量。由于这些单一的散射分量都对应于具有不同物理和几何特征以及分布特征的地物，从而提供了更加丰富的地表覆盖信息，这样就很大程度地改善了地物信息的分类精度；然后利用分解后单一散射分量数据结合传统的极化合成数据，可以得到更多的互不相关的数据源，再使用神经网络分类法对这些数据进行分类。分类结果表明，这种方法大幅度提高了全极化SAR数据用于实验区土地覆盖分类的精度。这种分类方法也可以广泛地用于SAR数据地表覆盖和土地利用动态监测和地表参数的提取。

Neural Networks Classification of Quad-polarization SAR Data Based on Target Decomposition ABSTRACT

SIR-C is the first spaceborne imaging Radar system with multi-wavelength and quad-polarization developed by joint effort of The U.S, Italy and Germany. Polarization SAR can measures the scattering matrix of each pixel on ground and synthesizes the image at given orientation and ellipticity angle , including linear and elliptical polarization. It has many advantages over single or multi-polarization SAR in detecting objects, identifying targets and extracting geometric structure of ground targets. During recent years, theoretical modeling and field experiments have established the fundamentals of active microwave remote sensing as an important tool in determining physical properties of ground objects. But different ground targets often have the same polarization signal characteristics because of the complexity of the distribution of the targets , which leads to wrong interpretation of the images and identification of the targets. Besides, relatively high correlation of the synthesized polarized images often lead to poor accuracy of classification. Based on SIR-C data of He Tian prefecture in Xinjiang of China, we use target decomposition theory to decompose the data into three no-related scattering components: an odd number of reflections, an even number reflections, and a cross-polarized scattering power, which represent different scattering mechanism of different objects. This decomposition technique allows us to obtain the estimation of single and double reflection components of backscattering coefficients for VV and HH polarization .They greatly improve the correctness of identification of ground objects. And what is more, the three components are non-correlated., which provides richer data resource. This paper employed neural networks classifier to classify the SAR images by combining them with polarimetric synthesized SAR power image. The decomposition result shows that the decomposed three scattering components reflect the correct scattering feature. The classification result shows that the method can effectively extract information of land cover, achieve the better classification accuracy of ground objects and improve the ability of SAR to monitor the land use and cover.