Rare endmembers estimation by NMF methods using multitemporal hyperspectral data and change informations

Autor:	Mireille Guillaume, C. Louis, G. Brisebarre
Rok vydání:	2014
Předmět:	Endmember Pixel business.industry Computer science Hyperspectral imaging Pattern recognition Artificial intelligence business Focus (optics) Change detection Field (computer science) Image (mathematics) Non-negative matrix factorization
Zdroj:	WHISPERS
DOI:	10.1109/whispers.2014.8077509
Popis:	Rare endmembers estimation is a very interesting and difficult issue in the unmixing field. We focus on the case of a rare endmember which appear as a change between two images of a same scene. We use both the information of the image where the new endmember is missing and change detection results to estimate the appearing endmember. We base the proposed approach on spectral unmixing with non negative matrix factorization (NMF), adding appropriate constraints on changed or non-changed pixels. We choose to test and discuss the alternate projected gradient optimization scheme. We compare the results to those of the estimation of appearing endmember through classical NMF unmixing on simulated data.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::df2bfb95882770c06964a1dcfd117fad https://doi.org/10.1109/whispers.2014.8077509 Zobrazit plný text záznamu