Semantic analysis of EO-1 hyperion hyperspectral data

Abstract

In this paper we propose a new method for unsupervised \ndata model discovery in hyperspectral data, by adopting the \nLatent Dirichlet Allocation (LDA) text modeling tool. The \nproposed method relies on defining a representation of \nhyperspectral data that allows LDA analysis, by defining a \ncorrespondence between hyperspectral profiles and visual \nwords. As such, we can use Latent Dirichlet Allocation as a \nmethod for discovery of latent (intrinsic and natural) \nmeaningful grouping of the spectral bands. Based on the \nmodel parameters we propose several use scenarios for \nhyperspectral data analysis, allowing us to identify \nsemantically meaningful structures in the observed scene. \nExperiments using the proposed method were carried on an \nEO-1 Hyperion data set imaging an agricultural area in \nRomania, acquired in June 2009.