5.7 Conclusions

Estimating the number of signal sources/signatures is a crucial preprocessing step in hyperspectral data exploitation. Its accuracy has significant impact on image analysis and interpretation. Recently, the concept of VD proposed in Chang (2003a) and Chang and Du (2004) has shown success in estimating the number of spectrally distinct signatures present in hyperspectral imagery. The exploration of VD was also investigated for many applications (Chang, 2006). This chapter revisits the concept of VD and redefines it as a general terminology to deal with signal sources of interest in various applications where the value of VD should vary with applications of interest and should not be fixed at a single constant value that fits all applications. In order to address this need, two types of criteria, data characterization-driven criteria and data representation-driven criteria, are used to develop VD estimation techniques where the former is focused on data characterization in terms of spectral statistics regardless of applications as opposed to the latter that finds an optimal number of signal sources to best represent data via a linear model. Tables 5.13 and 5.14 categorize these two types of criteria according to their design rationales. Category I includes factor analysis-based criteria, called real error (RE). Imbedded error (IE), extracted error (XE), and empirical indicator function (EIF) proposed by Malinowski (1997a, 1977b) and two criteria arising in passive ...

Get Hyperspectral Data Processing: Algorithm Design and Analysis now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.