Illumination invariant hyperspectral image unmixing based on a digital surface model

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Hyperspectral Imaging

Background:

• Spectral variability in hyperspectral data is a significant challenge for accurate analysis.
• Existing spectral unmixing models struggle to account for illumination variations and shadows.
• The physical mechanisms driving spectral variability under changing illumination remain poorly understood.

Purpose of the Study:

• To propose a novel spectral unmixing model, Illumination Invariant Spectral Unmixing (IISU), that physically addresses illumination and shadow effects.
• To leverage hyperspectral radiance data and LiDAR-derived Digital Surface Models (DSM) for improved unmixing accuracy.
• To provide a robust framework for endmember variability explanation from a radiance perspective.

Main Methods:

• Developed the Illumination Invariant Spectral Unmixing (IISU) model.
• Integrated hyperspectral radiance data with a LiDAR-derived Digital Surface Model (DSM).
• Utilized DSM-derived parameters (incident angles, sky factors, sun visibility) to model illumination variations.
• Employed a straightforward optimization procedure for efficient model solving.

Main Results:

• The IISU model demonstrated superior performance compared to state-of-the-art unmixing models, particularly in shaded pixels.
• Accurate estimation of endmember abundances was achieved by the proposed model.
• The IISU model successfully provided shadow-compensated reflectance, improving data interpretation.
• Physical explanation of endmember variability was supported by DSM-derived illumination parameters.

Conclusions:

• The proposed IISU model offers a significant advancement in spectral unmixing by physically incorporating illumination and shadow effects.
• The integration of hyperspectral radiance and LiDAR-derived DSM data provides a robust approach to address spectral variability.
• IISU enhances the accuracy of abundance estimation and reflectance retrieval in challenging illumination conditions, crucial for remote sensing applications.