Polarimetric SAR Surface Parameters Inversion Based on Neural Network
基于神经网络方法的极化雷达地表参数反演

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.