Noise reduction and best band selection techniques for improving classification results using hyperspectral data: application to lithological mapping in Canada's Arctic

Two problems in using hyperspectral data are the effects of noise and redundancy of information (too many channels) on classification results. This paper introduces two methods for dealing with these problems using a hyperspectral dataset over southern Baffin Island in Canada's Arctic region. This paper shows how classification results using matched filtering (MF) are improved on the modified datasets for identifying various lithologies. The noise-reduced dataset produced using the inverse minimum noise fraction (MNF) transform provides the most accurate classifications, followed by the dataset comprising the best bands determined through eigenvector analysis. The supervised approach presented in this paper in which training areas are extracted using a combination of visual analysis of MNF component images, analysis of existing geology, and field observations followed by match filtering produce useful spectral maps to assist in focusing field mapping activities or as stand-alone maps that provide lithologic information even in the absence of field mapping.