Work

Adaptive Endmember Bundle Extraction With a Modified Differential Evolution Algorithm for Addressing Spectral Variability in Geology-1 Hyperspectral Imagery

Year: 2026

Type: article

Abstract: High-precision extraction of alteration mineral information provides crucial guidance for hyperspectral remote sensing mineral exploration. Currently, widely used hyperspectral datasets (such as ZY-1 02D and GF-5) lack targeted design in spectral coverage and sensitive bands for minerals. The successful launch of China's first dedicated hyperspectral satellite for geology and mineral exploration - Geology-1 - has addressed this gap and provides an optimal data source for detailed alteration mineral identification. However, mineral spectra are susceptible to variation under complex geological conditions. To effectively characterize this spectral variability, the endmember bundle method is adopted, which depicts an endmember category through a set of spectra. Existing endmember bundle extraction methods suffer from low computational efficiency and inadequate accuracy when processing high-dimensional hyperspectral data under complex geological conditions, which limits the effective interpretation of mineral information. To address this challenge, we propose an Adaptive Differential Evolution Endmember Bundle Extraction method (ADE-EBE). Based on a multi-stage optimization framework, this method transforms high-dimensional discrete space into low-dimensional continuous space for dynamic optimization. By employing an adaptive differential evolution algorithm to globally search for optimal endmember combinations, it significantly enhances the stability and accuracy of endmember bundle extraction. Experiments on simulated dataset and Geology-1 hyperspectral imagery show that ADE-EBE achieves superior overall performance compared to traditional and state-of-the-art methods in terms of spectral fidelity, abundance accuracy, and computational efficiency.

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

DOI: 10.1109/jstars.2026.3703026

Authors: Linpei Yang, Ruyi Feng, Banxiao Ruan, Kaijun Yang

Institution: China University of Geosciences

Language: English

Topic: Remote-Sensing Image Classification

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $1,800

Funders: National Natural Science Foundation of China, Natural Science Foundation of Hubei Province