Abstract
To represent the spectral and spatial character of hyperspectral data, by introducing the smoothness constraint of hyperspectral data and the sparseness constraint of spatial distribution of the materials, an improved nonnegative matrix factorization (INMF) was used for hyperspectral unmixing. Its monotonic convergence is guaranteed by using a gradient-based optimization algorithm. Experiments demonstrate that the INMF algorithm is yielding accurate estimation of both endmember spectra and abundance maps.
| Original language | English |
|---|---|
| Pages (from-to) | 262-267 |
| Number of pages | 6 |
| Journal | Shenzhen Daxue Xuebao (Ligong Ban)/Journal of Shenzhen University Science and Engineering |
| Volume | 26 |
| Issue number | 3 |
| Publication status | Published - Jul 2009 |
| Externally published | Yes |
Keywords
- Blind source separation
- Hyperspectral unmixing
- Linear spectral mixing model
- Mixing pixel
- Nonnegative matrix factorization
ASJC Scopus subject areas
- Computer Science (miscellaneous)
- Engineering (miscellaneous)