Unsupervised signature extraction and separation in hyperspectral images: a noise-adjusted fast independent component analysis approach

Multispectral/hyperspectral imaging spectrometry in earth re- mote sensing applications mostly focuses on determining the identity and abundance of materials in a geographic area of interest. Without any prior knowledge, however, it is generally very difficult to identify and determine how many endmembers reside in a scene. We cope with this limitation by estimating the number of endmembers using a noise- adjusted version of the transformed Gerschgorin disk approach (NATGD). This estimated result is then applied to a noise-adjusted ver- sion of fast independent component analysis (NAFICA). Experimental results indicate that NAFICA offers a new approach for unsupervised signature extraction and separation in hyperspectral images.