Rice growth state estimation by hyperspectral manifold learning

Hyperspectral remote sensing is a promising method for the farm product monitoring. However, the estimation accuracy is restricted by the multidimensionality and shortage of statistically sufficient number of data. In this paper, a new method is proposed to acquire inherent vegetation-related coordinates on hyperspectral manifold by the combination of unsupervised manifold learning and supervised vegetation-related coordinates estimation. Experimental results show high estimation performance in vegetation-related quantities by the proposed method, i.e. nonlinear structure extraction and improved generalization performance, in comparison with multivariate linear regression based on hyperspectral data.