Minimum noise fraction transform for improving the classification of airborne hyperspectral data: two case studies

This paper investigates the use of Minimum Noise Fraction (MNF) components to improve the spectral separability of two specific thematic classes in airborne hyperspectral imagery using Spectral Angle Mapper (SAM). Particularly, we compared trends on data distribution before and after MNF transform. Two different data sets recorded with the Multispectral Infrared Visible Imaging Spectrometer (MIVIS) were analyzed. In the first case study, the classification of MNF-transformed data led to an overall enhancement in mapping asbestos roofs. In the second case study, the classification of MNF-transformed data succeeded to distinguish between two different artificial lakes, whereas classification of original hyperspectral data failed. Overall, this study showed how the use of MNF as pre-processing could improve the capability to extract information from two different airborne hyperspectral data sets.