33.2 Endemember Extraction

Endmember extraction has received considerable interests in recent years. This is mainly because with very high spectral resolution provided by hundreds of contiguous spectral bands hyperspectral sensors are capable of extracting many unknown and subtle signal sources which cannot be generally resolved by multispectral sensors. One of such signal sources is endmembers which provides crucial and vital information in data exploitation. Particularly, endmembers can be used to specify spectral classes present in the data. According to the definition in Schowengerdt (1997), an endmember is an idealized, pure signature for a class. So, technically speaking, an endmember is a signature generally available in a spectral library or database, and it is not necessarily a data sample vector present in the data. With this clarification, a pure data sample vector is specified by an endmember, but may not be the other around. Unfortunately, in real applications pure data sample vectors probably never exist because they may be contaminated by many unexpected effects resulting from noise, clutters, unknown interferers, etc. Accordingly, when an endmember extraction algorithm is applied to real data, it intends to find those data sample vectors which are most likely pure and represent endmembers. In unsupervised LSMA (ULSMA) discussed in Chapter 17, such data sample vectors are referred to as virtual signatures (VSs) and can be found by an unsupervised algorithm to be ...

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.