The noise-adjusted hybrid sub-pixel detector

Automatic detection of sub-pixel materials presents a challenging problem in hyperspectral signal processing. While a number of different approaches have been proposed in the literature, few of these can be considered fully self-contained. In most cases, research has focused on only one area of sub-pixel detection such as endmember extraction, parameter estimation or detector architecture. This paper presents a fully automated sub-pixel detection algorithm based on the hybrid sub-pixel detector, the iterative error analysis endmember extraction technique, and a Neyman-Pearson approach to tie desired performance to the required estimation of parameters. Results on real-world hyperspectral data show the improved performance of the proposed noise-adjusted hybrid sub-pixel detector (NAHSD) over conventional methods. Additionally, results demonstrate the ability of the NAHSD to achieve performance matching that attainable when sub-pixel detector parameters are perfectly known.