ABSTRACT
This chapter introduces hyperspectral data transform methods principal components analysis (PCA), signal-to-noise ratio (SNR)-based maximum noise fraction (MNF) transforms, and independent component analysis (ICA) and feature extraction methods, including canonical discriminant analysis and wavelet transform. It discusses a relatively complete collection of hyperspectral data analysis and processing techniques and methods. The goal of PCA is to find principal components in accordance with maximum variance in a hyperspectral image. The PCA, SNR-based principal component transforms are performed based on underlying second-order statistics of data, which implies an assumption that the random vector process underlying hyperspectral imagery is multivariate normal. The purpose of using the MNF transform process is translating the apparent surface reflectance data to have zero mean. The key idea of ICA assumes that data are linearly mixed by a set of separate independent signal sources and can then be used to unmix these signal sources according to their statistical independency measured by mutual information.
