Hyperspectral Data Processing: Algorithm Design and Analysis

11

Exploration on Relationships among Endmember Extraction Algorithms

Chapters 7–8 provide a series of studies on development and design of endmember extraction algorithms (EEAs) from an algorithmic implementation's point of view where Chapter 7 presents various simultaneous EEAs (SM-EEAs) that find all the desired endmembers simultaneously as opposed to sequential EEAs (SQ-EEAs) in Chapter 8 that find all the desired endmembers one after another in sequence. Due to the use of random initial endmembers SM-EEAs and SQ-EEAs produce final extracted endmembers that are inconsistent and not repeatable. To address this issue, two completely opposite approaches are investigated: initialization-driven EEAs (ID-EEAs) in Chapter 9 with initial endmembers generated by a custom-designed algorithm and random EEAs (REEAs) in Chapter 10 with random initial endmembers considered as realizations of a random algorithm. Despite that each of these four types of EEAs, that is, SM-EEAs, SQ-EEAs, ID-EEAs, and REEAs, is treated in individual chapters it is very interesting to investigate and explore their correlations and relationships. In particular, the two most widely used EEAs, PPI and N-FINDR, can be actually implemented in various versions derived from these four types of EEAs. This chapter is included for this purpose to further explore insights into the EEAs derived in Chapters 7–10.