Distributed Parallel Endmember Extraction of Hyperspectral Data Based on Spark

Algorithm 1

Endmember extraction based on IEA.

Input: Hyperspectral data , the number of endmembers m.

(1) Initialization: Threshold of (spectral angle), the number R for averaging the vectors

  with the largest error, and endmember matrix .

(2) Calculate the mean vector of the hyperspectral data .

(3) Perform the constrained unmixing on using as endmember matrix, and

   get the image of the errors (named “error image”) remaining after the unmixing.

repeat

(4) Find R pixel vectors with the largest error in the error image, and extract the subset of

   the set of R vectors, consisting of all those vectors which fall within the angle of the

   maximum error vector.

(5) Average the vectors in the subset to decrease the effects of outliers and noise, denoted

   as , and update endmember matrix .

until m endmembers are extracted.