An Augmented Linear Mixing Model to Address Spectral Variability for Hyperspectral Unmixing

Area of Science:

• Remote Sensing
• Signal Processing
• Data Science

Background:

• Hyperspectral imagery is crucial for material analysis but suffers from spectral variability.
• Traditional linear mixing models (LMM) struggle with spectral variability, hindering accurate abundance estimation.
• Accurate abundance maps are essential for various applications, including environmental monitoring and resource management.

Purpose of the Study:

• To develop a novel spectral mixture model, the augmented linear mixing model (ALMM), to effectively address spectral variability in hyperspectral unmixing.
• To improve the accuracy of abundance map estimation in the presence of complex spectral variations.
• To integrate a data-driven learning strategy for enhanced spectral unmixing performance.

Main Methods:

• Proposed the augmented linear mixing model (ALMM) to model spectral variability.
• Introduced a spectral variability dictionary to capture environmental and instrumental effects.
• Employed a dictionary learning technique with a low-coherence prior for simultaneous dictionary learning and abundance estimation.
• Validated the method using synthetic and real hyperspectral datasets.

Main Results:

• The ALMM effectively models spectral variability, including scaling factors and other environmental/instrumental effects.
• The embedded dictionary learning approach successfully learns the spectral variability dictionary.
• Experimental results demonstrate the superiority of ALMM over existing state-of-the-art methods in estimating abundance maps.
• Improved accuracy in abundance estimation was observed on both synthetic and real-world hyperspectral data.

Conclusions:

• The augmented linear mixing model (ALMM) provides a robust solution for hyperspectral unmixing challenges posed by spectral variability.
• The proposed data-driven dictionary learning strategy significantly enhances the accuracy of abundance map estimation.
• ALMM offers a promising advancement for hyperspectral data analysis, outperforming traditional methods.