面向极化SAR分类的地形辐射校正方法研究

In order to solve the problem of serious terrain effect in the classification of land cover types in complex terrain areas of polarimetric SAR (PolSAR) images, this paper analyzes the Polarisation Orientation Angle (POA), Effective Scattering Area (ESA) and Angular Variation Effect, AVE) correction and other terrain radiation correction methods are applied to the classification of PolSAR images, and an n-value determination method is proposed for AVE correction to improve the classification accuracy. Firstly, the SAR data after radiometric calibration and multi-look processing are corrected by POA based on circular polarization method, and the calibration results are geocoded. Secondly, ESA correction based on projection angle is carried out; then, the n value in AVE correction is determined based on the training samples, and then AVE correction is realized. Finally, GF-3 PolSAR images are used to carry out experiments, and the complex Wishart classifier is used for classification, and the classification accuracy is evaluated based on the verified samples. The experimental results show that the change before and after POA correction are not obvious. ESA correction can achieve the correction effect of about 3 dB, the percentage of standard deviation is reduced by 20.67%, and the overall classification accuracy is improved by about 9.42%. In addition, the AVE correction results were best obtained by selecting the n-value of the forest land most affected by topography, with a percentage reduction of 14.81% in the standard deviation, and an increase of about 8.2% in the ESA stage of the overall classification accuracy. The results show that there is little change before and after POA correction, and the effect of ESA correction is the most obvious. The AVE correction proposed in this paper can further effectively eliminate the influence of terrain effect on the basis of ESA.