8.8 Conclusions

SM-EEAs considered in Chapter 7 extract endmembers simultaneously for a given number of endmembers, p. One major difficulty of implementing SM-EEAs is finding all the endmembers at once, which results in high computational complexity. Another difficulty is that once the value of p is changed an SM-EEA must be reimplemented. In other words, all the previously generated endmembers by an SM-EEA cannot be used and new endmembers must be regenerated. Such a circumstance arises when p is not known precisely, and must be tested on a trial-and-error basis. This chapter has addressed these two issues by developing SQ-EEAs that implement SM-EEAs in a sequential manner. More specifically, an SQ-EEA finds one endmember at a time and one after another sequentially rather than all endmembers together simultaneously as an SM-EEA does. Most importantly, an SQ-EEA adapts its ability to numbers of endmembers, p, which can also be adaptive. It uses previously generated endmembers as part of future endmembers as the number of endmembers, p, grows. As a result, computational complexity can greatly be reduced. In addition, most SQ-EEAs do not require dimensionality reduction as SM-EEAs do. Finally, Table 8.1 summarizes design criteria of various SQ-EEAs presented in this chapter and their advantages and disadvantages.

Table 8.1 Summary of design criteria of SQ-EEAs and their advantages and disadvantages.

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.