A Review of Unsupervised Spectral Target Analysis for Hyperspectral Imagery

Abstract

One of great challenges in unsupervised hyperspectral target analysis is how to obtain desired knowledge in an unsupervised means directly from the data for image analysis. This paper provides a review of unsupervised target analysis by first addressing two fundamental issues, “what are material substances of interest, referred to as targets?” and “how can these targets be extracted from the data?” and then further developing least squares (LS)-based unsupervised algorithms for finding spectral targets for analysis. In order to validate and substantiate the proposed unsupervised hyperspectral target analysis, three applications in endmember extraction, target detection and linear spectral unmixing are considered where custom-designed synthetic images and real image scenes are used to conduct experiments.