Laplacian Eigenmap for Polarimetric SAR Image Classification

Inhalt:
In this paper the Laplacian Eigenmap (LE) technique is introduced into polarimetric SAR (PolSAR) feature dimensionality reduction and land cover classification. First, two categories of polarimetric features derived from original PolSAR data and polarimetric target decomposition respectively, as well as their combination, are considered as original feature sets. And then, the LE technique is utilized to map the original features in high dimensional space onto a low dimensional manifold to get intrinsic features. At last, classification is performed on these intrinsic features. In our experiments, KNN classifier is used to test the classification accuracy of the intrinsic features. The results show that by using the low dimensional intrinsic features instead of original high dimensional ones, better classification performance can be achieved.