The AMEE-PPI Method to Extract Typical Outcrop Endmembers from GF-5 Hyperspectral Images.

Area of Science:

• Remote Sensing
• Geoscience
• Image Analysis

Background:

• Mixed pixels are a significant challenge in hyperspectral endmember extraction.
• Existing methods often struggle with spatial context and user-defined parameters.

Purpose of the Study:

• To introduce AMEE-PPI, a novel hybrid method for robust endmember extraction.
• To evaluate AMEE-PPI's performance against established algorithms using real-world hyperspectral data.

Main Methods:

• AMEE-PPI integrates the Pure Pixel Index (PPI) with morphological operations (dilation/erosion).
• It uses HySime for noise estimation and signal-subspace inference to determine endmember count.
• Morphological elements of varying sizes (3x3 to 15x15) are applied to balance spatial information and local detail.

Main Results:

• AMEE-PPI achieved the lowest Spectral Angle Distance (SAD) and Spectral Information Divergence (SID) across different outcrop types.
• The method demonstrated superior similarity to field reference spectra compared to baseline algorithms.
• Visual analysis showed AMEE-PPI effectively avoided endmember leakage, producing cleaner spectral results.

Conclusions:

• Integrating spatial morphology with spectral purity enhances robustness in hyperspectral data analysis.
• AMEE-PPI offers significant improvements for downstream applications like unmixing, classification, and geological interpretation.
• The method is particularly effective for complex geological regions with varying illumination and mixing conditions.