Improvement of the MVC-NMF Problem Using Particle Swarm Optimization for Mineralogical Unmixing of Noisy Hyperspectral D
- PDF / 1,477,866 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 48 Downloads / 172 Views
RESEARCH ARTICLE
Improvement of the MVC-NMF Problem Using Particle Swarm Optimization for Mineralogical Unmixing of Noisy Hyperspectral Data Tohid Nouri1
•
Majid M. Oskouei2 • Behrooz Alizadeh3 • Paolo Gamba4 • Andrea Marinoni4
Received: 4 June 2018 / Accepted: 23 October 2018 Ó Indian Society of Remote Sensing 2018
Abstract The Hyperion data are broadly available for different parts of the world. Considering the spatial resolution of Hyperion (30 9 30 m2), it is rarely possible to find a pixel consisting of only one mineral. Spectral unmixing is therefore an important procedure in which dataset pixels are demixed into various constituents. Endmember determination is the key stage in spectral unmixing. The algorithms which are not depended on the existence of pure pixels in images are more efficient particularly when the spatial resolution is low (e.g., Hyperion data). On the other hand, the lower signal-to-noise ratio of Hyperion data is a disadvantage. Minimum volume-constrained nonnegative matrix factorization (MVC-NMF) is an appropriate non-pure pixel-based algorithm in low SNR conditions. Still, MVC-NMF is based on a gradient technique and is therefore problematic in the case of large amount of data. Particle swarm optimization (PSO) is a metaheuristic algorithm and computational simplicity is its main advantage. The minimum volume-constrained version of PSO (MVCPSO) was then investigated on Western Ardabil Hyperion data and the results were compared with MVC-NMF. To validate the accuracy of the results, 20 surface samples were collected and analyzed by spectrometry and X-ray diffraction (XRD). Measured spectra by analytical spectral devices Inc. FieldSpec were used to create a native spectral library. Native spectra as well as United States Geological Survey mineral spectral library were applied for identification of unknown endmembers spectra. The XRD results were implemented for quantitative validation of abundances maps of endmembers using Average Abundance Ratio. Keywords Spectral unmixing Hyperion Endmember estimation Non-pure pixel-based algorithms MVC-PSO
Introduction Due to the low spatial resolution of hyperspectral spaceborne sensors, a set of different minerals exist inside each pixel of a hyperspectral dataset. Therefore, for mineralogical applications of hyperspectral data analysis, the main & Tohid Nouri [email protected] 1
Faculty of Engineering, University of Mohaghegh Ardabili, Ardabil, Iran
2
Mining Engineering Faculty, Sahand University of Technology, Tabriz, Iran
3
Faculty of Basic Sciences, Sahand University of Technology, Tabriz, Iran
4
Telecommunications and Remote Sensing Laboratory, Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, Italy
need is to decompose pixels into constituent endmembers (Keshava 2003). Endmember determination is the first and most challenging part of the unmixing procedure and is very important to enhance the accuracy of the detected minerals (Miao and Qi 2007; Mozaffar et al. 2008; Cui et al. 2011;
Data Loading...
