Abstract
Classification of land cover types is one important application of polarimetric synthetic aperture radar (PolSAR) remote sensing. There are numerous features that can be extracted from PolSAR images. Among them, eigenvalues λi, entropy H, alpha angle α, and anisotropy A are effective and popular tools for the analysis and quantitative estimation of the physical parameters. Nevertheless, the speckle noise appearing in PolSAR images reduces the accuracy of image classification. Consequently, it should be filtered correctly. Generally, filtering PolSAR data generate biased estimates of λi/H/A/α parameters. In this paper, we studied the effects of bias compensation on supervised and unsupervised PolSAR image classification. We applied the asymptotic quasi maximum likelihood estimator AQ-MLE and Yahia/Aguili's bias compensation methods. To improve the classification accuracies, we demonstrated that bias compensation must be associated with speckle reduction. The combination of the span with biased parameters reduced the effects of bias but did not eliminate it totally. Simulated and real data were used for validation.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have