원문정보
초록
영어
Abundance estimation is an important step of quantitative analysis of hyperspectral remote sensing data. Due to physical interpretation, sum-to-one and non-negativity constraints are generally imposed on the abundances of materials. This paper presents a geometric approach to fully constrained linear spectral unmixing using variable endmember sets for the pixels. First, an improved method for selecting per-pixel candidate endmember set is presented, which is suitable for dealing with hyperspectral image with large number of endmembers. To determine the optimal per-pixel endmember set from the entire endmembers present in the hyperspectral scene, an iterative partially constrained geometric unmixing is then performed, in which subspace projection is used for fully constrained least square estimation. The performance of the resulting unmixing algorithm is evaluated by comparison with benchmark unmixing algorithm on synthetic and real hyperspectral data.
목차
1. Introduction
2. Linear Mixing Model and Geometric Abundance Estimation
2.1. Linear Mixing Model
2.2. Geometric Abundance Estimation
3. Methodology
3.1. Selection of Candidate Endmembers Set
3.2. Sub-Simplex Plane Projection
3.3. GESPVE Algorithm
4. Experimental Results and Analysis
4.1. Experiments on Synthetic Data
4.2. Experiments on Real Data
5. Conclusion
Acknowledgements
References