科研之友
微信
新浪微博
Facebook
分享链接
摘要
L-band fully polarimetric Synthetic Aperture Radar (SAR) data, acquired in April 2009 by the Japanese Advanced Land Observing Satellite over Tehran, were used to investigate the effect of various polarimetric descriptors on the accuracy of land cover classification. Particular attention was paid to evaluating the differences between polarimetric parameters in the image classification and understanding the type of information they contain. For this purpose, several polarimetric parameters were first derived from SAR complex analysis and target decomposition techniques including backscattering cross sections, ratios of scattering matrix elements, polarimetric coherencies, Pauli coefficients, Krogager coefficients, Freeman coefficients, Yamaguchi coefficients, H/A/Alpha parameters, and Eigenvalues of the coherency matrix. These parameters were then employed in Maximum Likelihood Classification (MLC), Minimum Distance Classification (MDC), and Parallelepiped classification for the land cover classification. The final accuracy assessment of the classification was performed using several statistical tests. This evaluation demonstrates that the performance and accuracy of the classification depend both on the selection of the polarimetric parameters and the type of classifier used. Polarimetric parameters extracted from target decomposition techniques in MLC produce the highest accuracies (a Kappa coefficient of greater than 90%) and the best discrimination of natural and man-made targets. Additionally, higher accuracies are generally obtained when the polarimetric parameters are derived from SAR complex analysis in a circular rather than linear basis. Thus, changing the polarization basis can significantly improve the result of classification.
- 出版日期2012-10
