基于神经网络方法的极化雷达地表参数反演

Polarimetric SAR Surface Parameters Inversion Based on Neural Network

人工神经网络（Artificial Neural Network）是一个由独立处理单元以一定拓扑结构高度连接而成的并行分布式信息处理结构,适于解决各种非线性问题。积分方程（Integrated Equation Model）单散射模型可模拟各种地表参数条件下裸露地表后向散射系数。以IEM为基础生成训练数据,用L波段和C波段SIR-C HH,VV极化单散射后向散射系数数据为神经网络输入,通过后向反馈（BP）神经网络模型可同时反演得到裸露地表条件下地表介电常数、地表相关长度和均方根高度等地表参数。

Surface parameters inversion by using Polarimetric SAR includes the inversion of the soil surface permeability , correlation length and RMS height. The retrieval of scattering parameters can be viewed as a mapping problem from the domain of measured signals to the range of surface/medium characteristics that quantify the observed medium. To date, parameter inversion has been based largely on empirical models. Empirical models have usually avoided the nonuniqueness problem by limiting the validity of the model to a single parameter and a narrow range. This limit on the range of validity requires that multiple empirical models be created-one model for each parameter. In this study, the Spaceborn Imaging Radar SIR-C data at L and C band was used to perform the inversion of bare surface parameters. A BP neural network based on IEM model was developed to carry out the inversion, and a test method was also developed. The combination of a scattering model (IEM) and NN makes it possible to perform inversion with higher accuracy and in real time. Backscat-tering coefficients computed from the model inverted surface parameters was proved to be good, compared with the real backscattering coefficients from radar image.