Loading...

Applications of Blind Source Separation(BSS) and Sparse Decomposition in Hyperspectral Image Processing

Zandifar, Azar | 2012

991 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 43198 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Babaiezadeh, Massoud; Ashtiani, Farid
  7. Abstract:
  8. Spectral Images, and Hyperspectral images as one of their main subsets, has been widely utilized in many scientific fields in recent years. Spectral unmixing may be regarded as one the main problems in hyperspectral image processing. Determining constituent materials (Endmembers) and their respective proportions(Abundance) is the main goal of spectral unmixing. Classic methods which are available for spectral unmixing mainly consist of two major separate steps for endmember extraction and abundance estimation. To combine these two steps in one, recently, powerful signal processing tools such as Independent Component Analysis (ICA), Nonnegative Matrix Factorization (NMF), and Sparse Component Analysis (SCA) have been applied to the field. According to above mentioned facts, in this thesis, we try to improve spectral unmixing algorithms to gain better quality of unmixing,by means of novel signal processing tools such as NMF, ICA and SCA. In this thesis, we introduced a new NMF-based algorithm with additional assumption on independence of different endmember’s abundances in each pixel, to unmix hyperspectral data. Comparing its performance with similar methods, we observed that this algorithm have relatively improved the unmixing quality and accuracy. In addition, we have applied Smoothed L0 norm (SL0) and Robust- SL0 algorithms adopting trivial modifications in these algorithms in accordance with the problem characteristics. According to comparisons with similar methods, we observe that SL0 shows acceptable result in unmixing hyperspectral simulated data in absence of noise, while performance of Robust-Sl0, the modified version of SL0 for noisy settings, decreases significantly in noisy environments
  9. Keywords:
  10. Independent Component Analysis (ICA) ; Smoothed L0 Norm (SLO)Algorithm ; Sparse Component Analysis (SCA) ; Spectral Unmixing ; Endmember ; Abundance ; Non-Negative Matrix Factorization (NMF)

 Digital Object List

 Bookmark

No TOC