ABSTRACT
This chapter discusses a deep understanding of the correct approaches for applying exploratory principal component analysis (PCA) on hyperspectral data, which is a fundamental step for every multivariate data processing protocol. Since PCA is applicable only on two-dimensional data matrices, hypercubes have to be reorganized in a bi-dimensional structure prior to their processing. When a hyperspectral image is submitted to multivariate data processing, including PCA, several approaches can be applied, depending not only on the type of data but also on the specific task that has to be achieved. In a multivariate system, such as hyperspectral images, it is fundamental to deal with methods that allow exploration of the data structure in a visual way understanding typical characteristics, in terms of sample groupings and variable intercorrelations. Exploratory methods belong to the family of unsupervised strategies in which a priori information about sample identity is neither required nor used for building models.
