Hyperspectral Data Processing: Algorithm Design and Analysis

10.3 Random VCA (RVCA)

VCA in Chapter 7 can be considered as a sequential version of PPI since it also uses OP as a criterion to find one endmember at a time by growing a series of convex hulls whose vertices can be used to identify endmembers. Because the initial endmembers used by VCA are generated by Gaussian random variables, VCA can be viewed as Gaussian VCA. However, according to experiments conducted in Chapter 7, the use of a Gaussian random variable is not crucial and can be replaced with any other random variable such as uniform random variable, which also serves as well as a Gaussian random variable does. In this case, we use a random generator to generate random initial conditions as do all the random EEAs to be designed in this chapter. The details of random VCA can be described as follows.

Random VCA Algorithm

1. Let VD to determine p. Set E⁽¹⁾ = 0 and .

2. Apply VCA to generate p random initial endmembers, denoted by .

3. If , let and go step 2. Otherwise, continue.

4. Find the intersection of . It should be noted that due to spectral variation in real data, a perfect ...