Construction of fast and robust N-FINDR algorithm

N-FINDR has been a popular algorithm of endmember (EM) extraction method for its fully automation and relative efficiency. Unfortunately, innumerable volume calculation leads to a low speed of the algorithm and so becomes a limitation to its applications. Additionally, the algorithm is vulnerable to outliers that widely exist in hyperspectral data. In this paper, distance measure is adopted in place of volume one to speed up the algorithm and outliers are effectively controlled to endow the algorithm with robustness. Experiments show the improved algorithm is very fast and robust.