Blind separation of spectral signatures in hyperspectral imagery

摘要

For the purpose of material identification, methods for exploringhyperspectral images with minimal human intervention have beeninvestigated. Without any prior knowledge, it is extremely difficult toidentify or determine how many endmembers in a scene. To tackle thisproblem, a new spectral unmixing technique, the spectral data explorer(SDE), is presented. SDE is a hybrid approach combining the optimalparts of fast independent component analysis (FastICA) andnoise-adjusted principal components analysis (NAPCA). Experimentalresults show that SDE is highly efficient for separating significantsignatures of hyperspectral images in a blind environment