Blind Hyperspectral Unmixing Using an Extended Linear Mixing Model to Address Spectral Variability.

Area of Science:

• Hyperspectral Imaging
• Remote Sensing
• Signal Processing

Background:

• Spectral unmixing is crucial for analyzing hyperspectral data.
• Linear Mixture Models (LMM) are common but assume static endmember spectra.
• Real-world factors like illumination cause spectral variability, limiting LMM accuracy.

Purpose of the Study:

• To develop an improved spectral unmixing algorithm for hyperspectral data.
• To address the challenge of pixelwise spectral variability in endmember signatures.
• To enhance the accuracy of material identification and abundance estimation in hyperspectral scenes.

Main Methods:

• Utilized a recently proposed extended Linear Mixture Model (LMM).
• Developed an algorithm allowing pixelwise, spatially coherent local variations of endmembers.
• Incorporated estimation of scaling factors to model endmember variability.

Main Results:

• The proposed algorithm demonstrated superior performance on synthetic and real hyperspectral datasets.
• Outperformed existing methods designed to handle spectral variability.
• Successfully provided accurate estimations of endmember variability across the scene via scaling factors.

Conclusions:

• The extended LMM framework effectively models and quantifies spectral variability.
• The developed algorithm offers a significant advancement in hyperspectral data analysis.
• Accurate estimation of spectral variability improves the reliability of hyperspectral unmixing.