Hyperspectral Imaging: an Emerging Technique in Remote Sensing

Hyperspectral imaging has expanded capability of multispectral imaging in many prospects. Specifically, it changes many ways how algorithms are designed from traditional spatial-domain (literal) analysis to spectral-domain (non-literal) analysis. This is mainly due to the fact that many unknown substances which cannot be identified by visual inspection or prior knowledge can now be uncovered by high spectral resolution provided by hyperspectral imagery. It is particularly evidential in applications of spectral unmixing, anomaly detection and subpixel/mixed pixel analysis where target signatures are on longer pure but rather mixed by other substances or background signatures within a single pixel. In this case, no inter-pixel spatial correlation is available to perform literal analysis and data processing must be carried out on a single pixel basis in an unsupervised manner without appealing for spatial information. This talk investigates issues arising in hyperspectral imaging from statistical signal processing view points and further explores applications in which multispectral imaging seems ineffective but hyperspectral imaging has found potential promise and great success.