End-member extraction could be considered as the most challenging stage of the spectral unmixing process. In this study, a new approach is proposed based on error analysis of Linear Spectral Mixture Model (LSMM) to extract optimal pure pixels. First a number of approximate end-members are identified visually or using N-Finder algorithm then LSMM is applied to identify pixels with proportions greater than one (over-shoots). Over-shoots are then replaced with initial end-members and the LSMM is performed again. This process is continued until reduction of the number of overshoot and under-shoot pixels below 5% of total image pixels. According to the results, the proposed end-member extraction approach satisfies this criterion within a few iterations (2 or 3 runs). The total numbers of under/over shoots are estimated 4.17% and 3.55% of total image pixels respectively for choosing the initial end-members visually and by means of N-Finder algorithm.

Improvement of linear spectral unmixing results using over-shoot pixels (case study: URMIA lake basin)

Niroumand Jadidi, Milad;
2015

Abstract

End-member extraction could be considered as the most challenging stage of the spectral unmixing process. In this study, a new approach is proposed based on error analysis of Linear Spectral Mixture Model (LSMM) to extract optimal pure pixels. First a number of approximate end-members are identified visually or using N-Finder algorithm then LSMM is applied to identify pixels with proportions greater than one (over-shoots). Over-shoots are then replaced with initial end-members and the LSMM is performed again. This process is continued until reduction of the number of overshoot and under-shoot pixels below 5% of total image pixels. According to the results, the proposed end-member extraction approach satisfies this criterion within a few iterations (2 or 3 runs). The total numbers of under/over shoots are estimated 4.17% and 3.55% of total image pixels respectively for choosing the initial end-members visually and by means of N-Finder algorithm.
978-1-4799-7929-5
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11582/312505
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact