SEMI-SUPERVISED MANIFOLD LEARNING OF TIME-SERIES HYPERSPECTRAL FOREST IMAGES

The accuracy of regression based on hyperspectral data is degraded by a restricted number of labeled data and the curse of dimensionality inherent in the high-dimensional feature space. In this paper, we propose two types of semi-supervised manifold learning methods for regression by a combination of supervised learning based on a small number of labeled data and unsupervised learning based on abundant unlabeled feature data. The regression and nonlinear manifold learning are realized by a kernelization of generalized eigenvalue problems. The proposed methods are applied to synthetic manifold learning problems and time-series hyperspectral leaf-scale images of oak trees.