Framelet-Based Sparse Unmixing of Hyperspectral Images

Area of Science:

• Remote Sensing
• Signal Processing
• Image Analysis

Background:

• Spectral unmixing is crucial for analyzing hyperspectral data, estimating material abundances within pixels.
• Semi-supervised methods leverage spectral libraries for improved unmixing accuracy.
• Existing methods face challenges with noise and achieving optimal sparsity.

Purpose of the Study:

• To propose a novel semi-supervised spectral unmixing model named framelet-based sparse unmixing (FSU).
• To enhance the anti-noise capability and overall performance of hyperspectral unmixing.
• To analyze the theoretical properties and algorithmic solutions for the proposed model.

Main Methods:

• Framelet decomposition to separate approximation and detail components of hyperspectral data.
• Promoting abundance sparsity in the framelet domain for improved representation.
• Utilizing the split Bregman algorithm for efficient model minimization and convergence.

Main Results:

• The FSU model demonstrates superior anti-noise capabilities due to framelet representation advantages.
• Experimental results on simulated and real hyperspectral data show improved unmixing performance compared to existing methods.
• The theoretical analysis confirms the existence and uniqueness of the minimizer for the FSU model.

Conclusions:

• The proposed framelet-based sparse unmixing (FSU) model offers a robust and effective approach for hyperspectral data analysis.
• FSU significantly improves unmixing accuracy and noise resilience.
• The method shows promise for various applications involving hyperspectral imaging.