A Novel Endmember Bundle Extraction Method Based on Modified Adaptive Differential Evolution Algorithm to Address Spectral Variability
Year: 2026
Type: conference-paper
Abstract: The accurate characterization of material composition from hyperspectral imagery is often hindered by spectral variability, a common phenomenon in complex environments. To address this, the endmember bundle extraction method represents a material class using a set of spectra rather than a fixed signature, thereby enhancing model flexibility. However, existing endmember bundle extraction methods typically suffer from high computational cost and limited accuracy when processing high-dimensional data under realistic, variable conditions. To overcome these limitations, we propose an Adaptive Differential Evolution Endmember Bundle Extraction (ADEBE) method. It employs a modified differential evolution algorithm to perform a global optimization for optimal endmember combinations, significantly improving the robustness and accuracy of endmember bundle extraction. Experiments on the widely used Urban benchmark dataset demonstrate that ADEBE achieves superior overall performance compared to traditional and state-of-the-art methods in terms of spectral fidelity, abundance estimation accuracy, and computational efficiency.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204603973