Extended Blind End-member and Abundance Estimation with Spatial Total Variation for Hyperspectral Imaging

Area of Science:

• Data Science
• Image Analysis
• Computational Imaging

Background:

• Blind linear unmixing (BLU) methods are essential for unsupervised separation of hyperspectral data into end-members and abundance maps.
• Noise in hyperspectral data often leads to granular and inaccurate abundance maps, limiting practical applications.

Purpose of the Study:

• To develop a novel BLU method that enhances spatial coherence in abundance estimations.
• To address the granularity issue in abundance maps caused by noise in hyperspectral imaging.

Main Methods:

• A new BLU formulation, EBEAE-STV (End-member and Abundance Extraction with Total Spatial Variation), was developed.
• The method incorporates spatial coherence into the cost function by adding internal abundances.
• A coordinate descent algorithm was employed to solve the proposed BLU formulation.

Main Results:

• The EBEAE-STV method significantly reduced granularity in estimated abundance maps using synthetic data.
• The proposed algorithm demonstrated lower estimation errors compared to existing state-of-the-art methodologies.
• Reduced computational times were observed for the EBEAE-STV method.

Conclusions:

• The EBEAE-STV methodology provides a robust approach for blind linear unmixing, improving spatial coherence in abundance maps.
• This method has potential clinical relevance for biomedical applications like tumor identification and tissue classification using hyperspectral imaging.