Supervised Locally Linear Embedding for polarimetric SAR image classification

In this paper the Supervised Locally Linear Embedding (SLLE) algorithm is introduced into polarimetric SAR (PolSAR) feature dimensionality reduction (DR) and land cover classification. SLLE technique, as a supervised nonlinear manifold learning method, can obtain a low-dimensional embedding space which preserves both the local geometric property of high-dimensional data and discriminative information from training samples. The combination of various polarimetric features are considered as original input feature sets. And then SLLE technique is implemented to map the original high dimensional space into a low dimensional manifold for subsequent classification. The experiments on fully polarimetric data show that SLLE can acquire more accurate classification results in comparison to many traditional DR techniques.