Using a New Search Strategy to Improve the Performance of N-FINDR Algorithm for End-Member Determination

• Published in: 2009 2nd International Congress on Image and Signal Processing pp. 1 - 4
• Main Authors: Ying Wang, Lei Guo, Nan Liang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2009
• Subjects: Algorithm design and analysis; Automation; Business; Data mining; Feature extraction; Hyperspectral imaging; Hyperspectral sensors; Iterative algorithms; Pixel; Stability
• ISBN: 1424441293; 9781424441297
• DOI: 10.1109/CISP.2009.5301014

A new search strategy proposed in this paper, which is used in n-dimensional spectral feature space to find reasonable end-members based on maximum volume transform (MVT), is implemented to improve the performance of N-FINDR algorithm. The N-FINDR algorithm, as a successfully used end-member extraction tool, produces inconsistent result in many cases and consumes lots of computing time if it carries out exhaustive search. In order to reduce the computation complexity and enhance the stability of N-FINDR, this search strategy is developed to eliminate most sample vectors, each of which is involved in computation unnecessarily, and select potential end-members sequentially. The new version N-FINDR algorithm with the search strategy presented in this paper will save computing time significantly and perform as well as original N-FINDR algorithm for end-member extraction, even better.