A Neyman-Pearson approach to estimating the number of endmembers

In this paper, we present a new method for estimating the number of endmembers for identification of materials in hyperspectral imagery. Unlike previous approaches, the proposed algorithm utilizes the Neyman-Pearson framework to estimate the number of endmembers that maximizes the probability of detection for a desired false alarm rate. The method can be used for any target signature, hyperspectral image, and detection algorithm. Results on multiple real-world hyperspectral images and material types will be provided showing the usefulness of the algorithm.